Materials and Metallurgy Subcommittee - September 21, 2000
UNITED STATES OF AMERICA
NUCLEAR REGULATORY COMMISSION
ADVISORY COMMITTEE ON REACTOR SAFEGUARDS
***
MEETING: MATERIALS AND METALLURGY SUBCOMMITTEE
USNRC
11545 Rockville Pike, Room T2-B3
Rockville, MD
Thursday, September 21, 2000
The committee met, pursuant to notice, at 8:30
a.m.
MEMBERS PRESENT:
GEORGE APOSTOLAKIS, Chairman, ACRS
THOMAS KRESS, Member, ACRS
WILLIAM SHACK, Member, ACRS
ROBERT SEALE, Member, ACRS
NOEL DUDLEY, Member, ACRS Staff. PARTICIPANTS:
E. HACKETT, RES
S. MALIK, RES
D. JACKSON, RES
L. ABRAMSON, RES
M. KIRK, RES
N. SIU, RES
H. WOODS, RES
D. BESSETTE, RES
D. KALINOUSKY, RES
T. DICKSON, ORNL
MODARRES, UNIV. OF MD.. BEGIN TAPE 1, SIDE 1:
[8:30 a.m.]
-- activities associated with PTS thermal
hydraulic experiments, flaw distribution, fracture toughness
distribution and model uncertainties, embrittlement
correlations, and the favored probabilistic fracture
mechanics code.
The subcommittee will gather information, analyze
relevant issues and facts, and 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 September 5,
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
they can be readily heard.
We have received no comments or requests for time
to make oral statements from members of the public.
The staff briefed this subcommittee on March 16
and April 27, 2000 concerning the status of the PTS
technical basis reevaluation project. At the May 2000 ACRS
meeting, the staff presented a draft Commission paper that
described potential options and approaches for revising the
PTS acceptance criteria.
Today we will hear presentations about the results
from some of the ongoing activities associated with the
reevaluation project.
We will now proceed and I call upon Mr. Edwin
Hackett, Assistant Chief of the Materials Engineering
Branch, to begin.
MR. HACKETT: Thank you, Mr. Chairman. Nothing
controversial in there. That's who I am. I guess this is
starting to get kind of comfortable for us.
We took this on as a major item. We took this on
as a major commitment to be briefing the committee on a
regular basis and we have been doing that. This is the
background. I think Bill mentioned this. There's been a
lot of encouraging developments on 99.
We show potential for significant burden
reduction, in a paper by Shaw Mallick and Terry Dixon, both
of whom are here. Additional developments. Thermal
hydraulics and PRA have occurred over the timeframe which
we've been looking at this.
And you'll hear about all of these pieces,
improvements in thermal hydraulic codes, testing at the APEX
facility for flow stagnation, which is ongoing, the context
for PRA, and explicit considerations of uncertainties.
I guess we're about a year and a half into it now.
The project has also been fully participatory. The original
plan completion, and I'll mention a little bit about this,
is December 2001. We're currently assessing some schedule
impacts. I think bottom line is we're behind that schedule,
but we're right now working on exactly how much.
Like I said, the project was fully participatory,
which is a pretty major departure for us from things that
we've done in the past, with input from key stakeholders,
and, obviously, within the NRC, that's principally the
Office of Research and the Office of Nuclear Reactor
Regulation, and our contractors.
The industry has been very active in this effort,
as we've talked about before, with the primary lead coming
from the materials reliability project for the PWRs, in
cooperation with EPRI, also, and the vendors. They have
provided probably close to half of the support for this
project in terms of some of what we'll come to here.
The plants that are participating, the
participation has really been all coordinated by the
industry and the MRP.
EPRI and the MRP have also been very key players
in the area of flaw density and distribution, in
volunteering materials and time and expertise in that area.
Debbie Jackson will probably talk about some of that in her
presentation.
The public is --
SPEAKER: Do they have their own codes, for
example, for the probabilistic fracture mechanics? Do they
use FAVOR, do they use VISA?
MR. HACKETT: They generally -- the agreement
previously was -- I guess I'd have to go back a ways. When
Mike Mayfield was branch chief, quite a while back, or even
section chief in Materials Engineering Branch, he had an
effort with Tim Griesbock, through ASME and EPRI, to
benchmark VISA at the time.
So I think the bottom line is that folks were
using VISA as kind of an industry NRC-wide view of looking
at this particular problem.
VISA basically evolved, of course, into what is
now FAVOR, and that's not quite fair to Terry. Favor is
much, much more than VISA was, but VISA and OKA-P, I think,
formed the bases for what became FAVOR, and Terry can talk
about some of that when he speaks later.
But I think there is pretty much consensus between
us and the industry that FAVOR is the code that will be
used. That's not to say there haven't been other codes that
people have applied. And, again, Terry could probably
address that better than I could.
But from Oak Ridge, there was OKA and OKA-P, I
believe, vintage mid-'80s, something like that. VISA, of
course, originated here with Ron Gambel and Jack Strosnider,
and then later versions through PNNL.
But the bottom line is that it's mostly been
standardized and I think there is agreement, the
NRC-industry working group, that FAVOR will be the code that
will be used for the probabilistic assessment, which is a
good thing.
It's like we don't need --
SPEAKER: You probably ought to check to see if
everything is still -- I say you probably ought to try to
check to see if there has been any divergence in that
expected uniformity and agreement.
MR. HACKETT: Right now, one of the things you
will here today is right now we've been focusing down on
getting some of these key inputs ready for delivery for
Terry to incorporate in the code.
So right now, everybody, including the industry
participants, is looking at getting a revised FAVOR that
basically we can start turning the crank on.
So that one has been there. I think the chairman
mentioned the reviews that have happened, starting with over
a year and a half ago and I think most recently with full
committee in May. I think Mark Cunningham and I were here
in both March and May to talk about the risk acceptance
criteria and other pieces.
You may remember that are four full-scale plants
being analyzed, again, with an awful lot of help from the
industry participants, and they are listed here.
The major deviation, which occurred basically
earlier this year, was H.B. Robinson dropped out of
participating in the project formally and Beaver Valley 1
agreed to replace them, basically, and that's gone pretty
well, but there were some delays associated with that.
Palisades has been a participant in this from the beginning.
The three IPTS plants, the integrated PTS
assessments that were done in the 1980s were Oconee, Calvert
Cliffs and Robinson. So we wanted to basically try to redo
those, and we did lose Robinson along the way, but we think
we've made up for that and we've made up some ground there.
And this is kind of where we are overall. This is
just big picture, and, again, you're going to hear an awful
lot more detail the rest of the day.
But the work is progressing in the major technical
areas pretty well. Sometimes we're an awful lot in the mode
of one step forward, two steps back, and trying to
recalibrate, and there are schedule issues, at least one of
which was related to Robinson dropping out and picking up
Beaver Valley, but there are other areas that are taking us
longer. It is a fairly ambitious undertaking overall.
I did mention this piece here, though. The
finalization of the materials inputs, we were hoping to have
completed that earlier this year. It looks like right now
we're hopefully on track for October-November, finalizing
things like the statistical distribution of the fracture
toughness, the embrittlement correlations, the flow density
and distribution, so we can get those to Terry Dixon and
others for incorporation into FAVOR.
It's a very interesting piece here that Farouk
Eltawila and Dave Bissette and others are working on with
validation of some of the thermal hydraulics work at the
APEX facility has been basically reconfigured to simulate
the Palisades plant.
Experiments are underway there. As a matter of
fact, one has been completed, and I believe there are seven
more that are anticipated between now and approximately the
end of the calendar year. So that's a major step.
Obviously, we're looking at conditions for flow stagnation
and mixing.
Progress in the PRA aspects, I think, was covered
also by the chairman. A big part of this project, of
course, that we stress every time is a process for explicit
consideration of uncertainties. We've never really done
that before. This has always been done as more of a
bounding thing.
There was a Commission paper completed which Mark
Cunningham had the lead for on the acceptance criteria in
July 2000. I think based on some comments from the
committee, that was recast to the Commission as an
information paper as opposed to asking for their specific
input at this point.
DR. KRESS: Can I ask you a question about the
uncertainties?
MR. HACKETT: Sure.
DR. KRESS: What do you plan on doing with those
when you get them?
MR. HACKETT: Well, overall, we, for the first
time, are looking at doing explicit uncertainty analyses of
each of the inputs and hoping to cascade that through.
Now, one of the things that's caused us some
hesitation or some angst over this is where we're going to
end up with that, and I think it's fairly --
DR. KRESS: That's actually my question. When you
get there, what are you going to do with it?
MR. HACKETT: When you get there, what we're not
going to do, I think I can say, is we're not going to line
up all the worst case scenarios, like we have before, and it
looks like maybe Nathan wants to make a comment on that.
But the intent was to do this -- I'll just say,
and then let Nathan get into some details, the intent was to
keep this as a, quote-unquote, best estimate analysis. It's
not supposed to be a bounding analysis by just the way it's
written in 10 CFR 5061.
So the intent was to try to keep this as a best
estimate and not go to a bounding case.
MR. SU: This is Nathan Su, Office of Research.
That's a great question. Part of the answer is we don't
know until we start seeing what the results start looking
like. If the results look like, for example, we've been
very conservative in the past, then we may not have to do a
whole lot with the calculation, just say, okay, we know how
to calculate the mean very well, here it is and work with
that.
Of course, we'll have a sense of the uncertainty
about that mean. If we're closer to -- if the risk is
higher than we think it is at this point, then we'll have to
do something about that.
So you're asking the general question what do you
do with the distribution once you've generated it and --
DR. KRESS: I liked one of your answers, and that
is that's one way, in fact, probably the only way to know
you've got a real mean.
MR. SU: Yes. And we're certainly going to do our
best to calculate that. But in terms of using the full
distribution, I guess we haven't really worked that out.
DR. KRESS: I was hoping it might have something
to do with questions of defense-in-depth and risk acceptance
criteria, but that's sort of another subject.
MR. HACKETT: That obviously needs to be factored
in, in a big way, and that kind of brings us to the next
point anyway, because we did get a fair bit of good dialogue
here with the committee on the risk acceptance criteria
through the ACRS meetings in March and May.
Basically, what was discussed a lot at those
meetings and also in the paper that Nathan and Mark and
others put together is a risk approach that's,
quote-unquote, similar to what's contained in 1.174, which
would obviously include explicit consideration of
defense-in-depth and other factors.
But it also has the effect, of course, in this
case of resetting a risk criterion that was set in, I guess
it's fair to say, kind of an ad hoc way originally at
5E-minus-six, has the effect probably of starting that
baseline at 1E-minus-six, and arguing from there one way or
the other as to which way this is going to go.
DR. APOSTOLAKIS: What is defense-in-depth in this
case?
DR. KRESS: That's a very good question, George.
I think it has to do with inspection and looking at coupons
and monitoring and that sort of thing.
MR. HACKETT: Inspection is an element. I think
one thing I would say, too, and I don't know how much --
it's a very good question. I don't know how much this is
actually defense-in-depth, but basically one of the things I
would say for this project is that you're looking at a
reactor vessel where you're assuming initiation of flaws
leading to through-wall failure, which is leading to a big
hole in the vessel, which is then likely going to be a
pretty major event for the containment to deal with.
So you're making those assumptions and maybe you
could say there is some aspect of that that involves some
defense-in-depth, whether all of that actually happens.
We know, for instance, that you can initiate a
crack and then arrest it. So you might arrest the crack.
Or you could have a crack that goes through-wall, and Mike
Mayfield would probably want to shoot me, but it may not go
to this foot-wide, 13-foot long thing, maybe it doesn't.
But the problem with saying that is I don't think
there are any of us who could quantify that. It's beyond
the state-of-the-art in fracture mechanics.
SPEAKER: Dan Marzinski told me he sees is as you
just get leak before break.
SPEAKER: It strikes me that sometimes it's
worthwhile to go back and recognize that there was
enlightenment before RG-1.174. You know, we've kind of
gotten jaded, I think, in our appraisal of what
ten-to-the-minus-six means, because generally in the context
of the application of 1.174, where you're comparing between
two alternatives, which is one of the cases that 1.174 was
set up for, you're still talking about relatively
controllable consequences.
By that, I mean, sure, you had a core that went to
hell and breakfast with TMI, but it didn't do the Chernobyl
thing, if you will.
But if you go back far enough, you recognize, I
think, that there were two categories of concern. One was
the higher risk event; that is, the risk numbers were in the
ten-to-the-minus-four and higher numbers. But the other was
where the consequences were in the extraordinarily severe
range.
I wonder if we're being very smart if we allow
ourselves to think in terms of ten-to-the-minus-six with
those extraordinarily severe consequence events.
It sounds to me that we're almost setting
ourselves up for that. We're selling ourselves a bill of
goods if we're not careful.
So I guess I'm going to be the small mind that's
going to provide the refuge for this idea, to paraphrase our
chairman, but I worry.
DR. KRESS: Let me ask you another question about
that, along the same lines. Is it considered by you guys
that if you have a PTS event that fails the vessel, you also
fail containment? Is this a LERF, as well as a CDF at the
same time?
MR. HACKETT: That's why I put up -- and I have
that on the last slide -- that's why I put up that last line
there, because when we briefed the committee earlier, and I
remember Dr. Kress and also Tom King was here, there was a
pretty good discussion that ensued over that.
I think the materials perspective, myself, Mike
Mayfield, others like that, the answer would be yes, we
think there would be some violation of containment
somewhere.
I think if Mike were here, he's almost of the mind
that he thinks it's almost for sure that -- and he's not
coming from the standpoint of even pressurization of the
containment.
He's saying that you now have this big jet force
that you put a big hole in one side of the vessel and you're
really pushing water and steam out and the vessel is
designed to radially expand anyway on the supports.
So you would slam the vessel into one side of the
shield wall. You'd be into some plant specifics about gaps
and so on, and that, of course, is going to drag along with
it the other pieces of the primary and it would almost be
naive to think that at some point, some containment
penetration isn't going to be pulled loose or something.
DR. KRESS: Do you have estimates for those forces
and things or has that been part of the problem?
MR. HACKETT: Dave Bissette was discussing this
with us yesterday. The short answer is no, I don't believe,
but there are estimates for things like hot leg and cold leg
breaks, and I suppose jet force is associated with those.
I'm obviously out of my depth here. Dave might be
able to address some of that.
But I think the bottom line is the expectation
would be that it would be enough to move the primary in a
significant way. But to the present the other point of
view, I think if Mark Cunningham were here, I think Mark was
looking at it as a -- trying to bound the problem.
If you were to be able to take a subset, well, I
only have X number of plants that I think I have a PTS
problem with anyway and let's say it just happens to be
they're all large dry containments and maybe I've got
sliding supports on the generators and things aren't as bad
as I've just described, for whatever reason, that maybe you
could make that argument.
And that's, I believe, where the committee was
coming from, that, well, at least you could consider some
arguments about containment integrity to set this criterion
5E-minus-six or even lower, if you could argue convincingly
that your container was so robust.
The problem, I guess, that we see is that making
that argument, I think, would be a very difficult thing for
a licensee to do.
They would probably -- if I were a licensee, I
think I'd say to Mr. NRC, I'd like to see a reg guide on how
to come argue with you about my containment integrity, and
then we're off into another multi-year effort of trying to
define that.
DR. KRESS: One of my interests is in risk
acceptance criteria and I'm very interested in this
one-times-ten-to-the-minus-six. My interpretation is that
the Reg Guide 1.174, LERF is
one-times-ten-to-the-minus-five, and this is one set of
sequences and you don't want them to add the whole thing in,
so a factor of ten is a good idea, maybe.
So that's where the one-times-ten-to-the-minus-six
comes in.
But the question I have about that, and I think
the committee will recognize where this is coming from, it
seems to me that when you have a PTS event, that what it
suddenly turns into is an ire ingression accident. The
steam is not there. It's ire coming through the openings
and naturally convecting.
Ire ingression accidents are quite different than
steam ingression accidents and that causes me to pause when
I look at the ten-to-the-minus-six as the criterion, because
that's based on steam oxidation driven core melt accidents.
MR. HACKETT: Right.
DR. KRESS: So I just wanted to point out I think
that's where that's coming from and I have a little bit of a
concern about that.
MR. HACKETT: That's a good point and it's not
really one we've considered. Good point.
This, I'll go ahead and not take up too much of
the time myself here, because we did cover it. Dr. Kress
mentioned LERF and I have that on here just in terms of
summary and conclusions, but this, for us, obviously, I
think, is the first application of sort of the new NRC
risk-informed methodology to revise, and we've been talking
around this, but what basically is an adequate protection
rule, which kind of puts us in an interesting space
philosophically, I think as Dr. Kress has been pointing out.
I think the progress has been good. We've tried
this before. This is the project that's kind of defied my
boss since I've known him. It's frustrated Mike for years
and years, Mike Mayfield, and I think he was the driving
force behind getting this going. So it's been going about
as good as it ever has here and it's a lot of credit to Mike
for that.
The consideration of LERF and containment
integrity is a major departure from what we've done before
in this area, but I think it's incumbent on us to, in this
environment we find ourselves in, it's incumbent on us to
consider these aspects.
I don't think it was something anyone went into
thinking that we're going to have a lot of fun doing this
maybe, but it's a valid thing to consider in the current
framework. And the old rule does not, obviously, get into
those kinds of considerations at all.
Another interesting piece is that this project was
basically marketed or sold as a licensee burden reduction
type of project, but I would say right now it's very much
complex enough that the final outcome is not entirely clear.
Dr. Kress pointed out the insertion of the
uncertainties and any kind of cascading effects, where what
we're building up to right now, and maybe Terry will talk
about some of this, is by the project schedule, we have an
initial scoping study run for Oconee that's scheduled to
complete somewhere in the December timeframe, which will
hopefully give us an idea of which way this vector is going.
We are hoping, obviously, that we are looking at a
relaxation of the current PTS criteria, but I think right
now it's fair to say it's probably too early to tell.
So that's basically what we're hoping to get an
indication of in the bottom piece. I think as far as the
future goes, I don't remember the exact schedule, but we
would probably be on the hook for coming back and having
further discussions with the committee by about the turn of
the calendar year, and we are down for a Commission paper, I
think, Shaw, in February of next year, that's going to be
addressing progress.
Hopefully, this type of piece would have been
considered by then, but we're a good ways away from that
right now, so that we have some schedule impacts to address.
But we're hoping that the next time we come
forward, we'll actually have some results from incorporating
all this good science and so on into what's actually a
probabilistic run for the first time.
So we're getting there. We're not exactly -- we
were hoping to be there kind of about now, but we are behind
in that schedule.
So I guess I could take any overall questions, or
otherwise we'll go into kind of a rundown of the three major
technical areas.
I guess, not hearing any, it looks like what we --
if we go in order here, I guess Roy Woods was going to come
on up and talk about some details of the PRA aspects.
SPEAKER: We'll just note that -- is
ten-to-the-minus-six an adequate protection rule or an
improved safety rule and will we have to backfit to get to
that level?
DR. APOSTOLAKIS: Right now, it's
five-ten-to-the-minus-six, is it not?
DR. KRESS: Now you're going down to one and my
contention --
DR. APOSTOLAKIS: And that would lead to further
reduction.
DR. KRESS: No, that's going the other way.
DR. APOSTOLAKIS: I'm confused, because I thought
I heard that --
MR. HACKETT: The intent was -- the thought was
there was enough conservatism in the way that you calculated
the screening criteria that was used to assure you got to
the five-times-ten-to-the-minus-six.
If you removed that conservatism, you would get
burden reduction. If you lower the criteria, even though
you still have conservatism, you're going both ways now and
it's not clear.
If you kept it at five-times-ten-to-the-minus-six,
I don't think there would be much question it would probably
reduce some burden.
DR. KRESS: Yes, but there was no real technical
basis for the five-times-ten-to-the-minus-six and I think
they were searching for --
DR. APOSTOLAKIS: That would depend a lot on the
containment.
DR. KRESS: It certainly would, in my mind, yes.
DR. APOSTOLAKIS: And I think the definition that
the Commission was giving to defense-in-depth and they talk
about multiple barriers, there is an implication of
redundancy. Otherwise, it wouldn't be multiple.
DR. KRESS: Absolutely.
DR. APOSTOLAKIS: So say that if the uncertainties
are very large, we're going to inspect such things, so why
defense-in-depth.
DR. KRESS: No, but a lot of people consider that
as one element of defense-in-depth. It's not your classic
defense-in-depth.
DR. APOSTOLAKIS: It's something you have to do.
Is the containment defense-in-depth? I don't know.
DR. KRESS: In this case, it may not be
defense-in-depth, because what we heard is that when you
have a PTS event, you're likely to fail containment.
So in an event that it fails both at the same
time, then the containment is not defense-in-depth for that
event.
DR. APOSTOLAKIS: Or if you need it to contain the
accident consequence, that is not defense-in-depth.
DR. KRESS: That's right.
DR. APOSTOLAKIS: It's not redundant.
DR. KRESS: No, I don't think you can --
DR. APOSTOLAKIS: You came up with some ideas like
that. Remember that? You were very happy that day. Now it
comes back to you.
DR. KRESS: You got to be careful what you say
around here.
MR. WOODS: Good morning. I'm Roy Woods. With me
is Nathan Su. We're with the Office of Nuclear Regulatory
Research, PRA Branch. Also with us Eric Thornsbury, at the
table. It's got a lot of the background information that
you will be able to tell, if you've asked that kind of
question, by the mad shuffling through the paper over in the
corner there.
I also want to point out that we did go through
and kind of practiced for this and I'm aware that there's a
lot of material to cover in my talk and the next two, and so
I'm going to try to hurry through the stuff you've already
heard before. If I go too fast, you will, of course, stop
me, please.
What we're trying to do, we're trying to basically
support the development of the technical basis for the
revised PTS rule and in order to do that, we're trying to
ensure that it's a coherent risk-informed process, with
appropriate integration of thermal hydraulics, PRA and
fracture mechanics.
There is a slide to follow that you've seen before
that shows that as a picture. We're also trying to make
sure we have a consistent treatment of uncertainties.
DR. APOSTOLAKIS: Roy, speaking of uncertainties,
is the paper from Maryland part of today's discussion?
SPEAKER: We didn't include that in the schedule.
DR. APOSTOLAKIS: When will it be discussed?
Because I have a lot of questions.
MR. WOODS: I thought Mohammed was going to be
here, but --
SPEAKER: He was going to listen in, but we didn't
have that scheduled. I know Ed Shaw --
MR. HACKETT: This is Ed Hackett. I guess what
we'll do is we'll take an action, Professor, to make that.
We might need to make that the subject of another meeting,
but we --
DR. APOSTOLAKIS: I think we should, because I'm
not sure I understand everything that is being said there,
and, in some instances, I'm not sure I agree, and this seems
to be a very important part of FAVOR because it -- I mean,
it characterizes the uncertainties and then propagates them
and so on and I thought we were going to discuss it today.
SPEAKER: Although, just to clarify, the status at
the moment is that the FAVOR works strictly on the
statistical correlations at the moment, right?
SPEAKER: I guess I need to probably here more
about that, Bill. Just statistical as opposed to
mechanistic?
SPEAKER: Whether the treatment of uncertainties
that are given in the Maryland paper, they're using the
statistical correlations that were developed at Oak Ridge.
SPEAKER: That's my understanding.
DR. APOSTOLAKIS: No. They go beyond that. They
provide --
SPEAKER: But, I mean, the calculation is actually
not using that at the moment, I don't think.
DR. APOSTOLAKIS: Oh, I see. So maybe there will
be time here for us to discuss it before the --
SPEAKER: Well, FAVOR, and Terry, I'm sure, will
talk to this, is going to incorporate the uncertainties in
the different ways that we said in the white paper back in,
I don't remember when that was issued, June or September,
something like that, which is -- so in particular, we're
deal with the aliatory uncertainties in the K1C and K1A
terms and the epistemic uncertainties and all the other
terms.
So FAVOR is being set up to address that. Now,
what specific distributions are going to be input to FAVOR
is the point of what Maryland is doing and you're right, we
haven't spoken about that to the committee.
DR. APOSTOLAKIS: But before you guys invest
significant amounts of effort in here, I think we ought to
have a meeting, because I'm not sure -- you shouldn't take
my comments as an ominous sign that there is major
disagreement, but I just don't know right now.
And by reading the paper, I get more confused than
I was before I started and I don't like that.
SPEAKER: Actually, I'm confused about that, too.
DR. APOSTOLAKIS: It's tough going, I'll tell you,
and I'm no sure I agree with the calculation scheme that is
proposed and given the emphasis that you guys have placed on
uncertainties and consistent treatment and so on, I don't
see it there.
SPEAKER: That's just self-defense. We keep
beating them over the head with uncertainties. They've got
to do some treatment of it.
DR. APOSTOLAKIS: Yes.
SPEAKER: But I went back to read Nathan's white
paper and it seemed to me that the way FAVOR now treats the
K1 distribution is purely aliatory.
DR. APOSTOLAKIS: And it shouldn't be.
SPEAKER: And it shouldn't be. Maryland is an
attempt to go the other way, but I got confused as to --
just to get off the subject a little bit. Somehow I would
pick those K1 curves. I see a family of curves in between
there.
SPEAKER: Yes, and -- you're right and --
SPEAKER: You would pick one of those curves and
that's really an epistemic, because I don't know which of
those curves to pick. But once I pick a curve, I'm
following along that curve. I'm not walking up and down
that whole distribution.
SPEAKER: Right, right, right.
SPEAKER: And FAVOR now doesn't do that, as I
understand it.
SPEAKER: I guess maybe -- I don't know, Terry, if
you were planning -- I didn't think we were going to get
into that depth in this particular presentation, even though
--
SPEAKER: Actually, it used to take a curve and --
SPEAKER: That's because you picked a lower bound
curve.
MR. DIXON: I'm Terry Dixon, from Oak Ridge
National Laboratory. The way that -- before the University
of Maryland got into being our advisor on this, we did, in
fact, pick one curve and we sampled from a Galsion
distribution to determine which one of those curves and then
we followed that curve down through the cool-down.
Now, as you said, Dr. Shack, we don't do that. We
actually, at a given moment in time or, in other words, a
particular T-minus-RTNDT, we are dealing with the
distribution at that vertical slice through T-minus-RTIT.
DR. APOSTOLAKIS: So you've selected already a
curve, but it would be epistemic, because you are selecting
--
SPEAKER: No. It seems to me purely aliatory.
DR. APOSTOLAKIS: Aliatory, yes.
SPEAKER: The idea or at least -- and Professor
Maderas can speak to this better than I can -- doing this
method introduces the aliatory uncertainty.
SPEAKER: I would have thought that you would have
had a curve with a small scatter band around it to take care
of the aliatory part, but to treat the whole scatter as
aliatory seems to me to be incorrect.
DR. APOSTOLAKIS: I think we're going to get into
this, but I really think -- in fact, let me ask you. Is it
possible to have a discussion here where you will walk us
through a detailed calculation based on figure six of the
Maryland paper?
SPEAKER: I intend to this afternoon in my
presentation.
DR. APOSTOLAKIS: Today. Well, I won't be here
this afternoon, but this is for -- I mean, I want a
detailed, how do you pick things, then what do you keep
track of. Do you really start by selecting a vessel? What
does that mean?
DR. KRESS: That's just a figure of speech.
DR. APOSTOLAKIS: Yes, but there is a distribution
and so on. No, but I really would like, because there are
two uncertainties here that we have to keep track of.
SPEAKER: Well, I like Nathan's paper, where you
have an epistemic loop and an aliatory loop, and I'd like to
know what's in the epistemic loop and what's the aliatory
loop.
SPEAKER: Yes.
DR. APOSTOLAKIS: So how does figure six in the
Maryland paper --
SPEAKER: I think we need to walk you through that
and, again, as Tom pointed out, this is a nomenclature,
picking a vessel, to fix the epistemic parameters. But,
again, we --
DR. APOSTOLAKIS: That's not my problem.
SPEAKER: I know.
DR. APOSTOLAKIS: But following the loops, I think
that's --
SPEAKER: Right, right. And this has been a point
of discussion among us for a while, trying to make sure we
got it right, and we do need to talk with you about that.
SPEAKER: Some have actually floated back and
forth in FAVOR.
SPEAKER: Yes. And I intend to talk at some level
of detail this afternoon about this.
SPEAKER: And conceptually, again, we had every
intention of addressing the epistemic uncertainties in the
aliatory distribution. Now, whether we're doing it right,
that's worth discussing.
MR. HACKETT: I think what we can commit to do --
this is Ed Hackett, of Research, again. We'll take an
action to make that happen, because I think that would be a
very useful thing to do.
I guess I would also just mention that later
today, Mark Kirk will be giving a presentation, wherein he
was going to at least attempt to cover conceptually the
breakout, aliatory and epistemic, in the statistical
evaluation of fracture toughness, because that -- I think
the committee is absolutely right.
That has not -- model uncertainty has not been
addressed in that area before and we're attempting to do
that now for the first time. I think what's been in there
has been pretty much all aliatory so far.
So we'll take an action to address that
separately, but maybe some of what Mark will talk about this
afternoon will at least try and conceptualize.
DR. APOSTOLAKIS: So it's possible to go through
one loop, the calculational loop, that would be extremely
useful.
MR. WOODS: Okay. I'm going to continue on with
the second bullet here then and I'm going to skip over most
of it. I thought we'd get hung up on the second one, but
that already happened with Ed.
Obviously, we're trying to develop a new screening
criteria and it will be based on something like RTP or TS,
embrittlement parameter, and also on the figures of merit,
which would be CDF, and maybe LERF, and also what the
acceptance criteria for the CDF or the LERF value would be,
which is kind of a separate thing that we weren't prepared
to talk about today.
DR. APOSTOLAKIS: Incidentally, on the previous subject,
Bill mentioned Nathan's paper and I also -- I read it some
time ago, but I also read the Maryland paper and the paper
by Dixon and Mallick.
Is it possible, in the future, that you guys make
sure you refer to each other, cite each other, and make sure
that the stuff is consistent, instead of throwing in a
reference, Nathan Su, and then we don't see any connection
to Nathan Su.
It would be very useful, in other words, if these
things are coming from the same project, to have some
consistency. That's not a major thing, but it's knowing.
SPEAKER: I think what you're saying, it's partly
a function of -- you know, we're hot in the development
process.
DR. APOSTOLAKIS: Right.
SPEAKER: We are certainly intending to document
how we deal with uncertainty in PTS in a specific report
that will address that, and it will basically, as I see it
right now, be an expansion of the white paper.
So we'll talk about how we're dealing with it in
thermal hydraulics, how we're dealing with it in PFM, how
we're dealing with it in HRA and so forth, and put it all
under this consistent framework.
But I guess we didn't think about doing that early
on, but, yes, you're right. We're holding meetings and
talking, but we're not necessarily documenting that in what
you see.
DR. KRESS: The risk acceptance criteria have been
worked on by the Risk Analysis Group rather than the group
you guys are in. Is that a different, sort of a separate
project?
SPEAKER: They are us, yes.
DR. KRESS: They are us. Okay.
MR. WOODS: Okay. Well, the last bullet I think
everybody is probably aware of. We started with the IPTS,
the plants that Ed mentioned. We're trying to reflect
changes to those plants. In fact, one of the plants itself
changed, Beaver Valley instead of H.B. Robinson.
Also, the very last thing on that slide, we
obviously have to get our arms around the risk from all the
plants, based on the analysis of four plants.
I'm going to just show this next one, but I think
everybody has seen it. This is the basic framework. You
start with identifying the PTS event scenarios that you're
worried about with a fairly standard PRA.
I'm going to show you an event tree in a minute,
and that defines which thermal hydraulic analysis you need.
You'll group certain events into a group and use one thermal
hydraulic analysis for all those events.
And the ultimate objective is to do the
probabilistic fracture mechanics and what you're showing
here is you're not certain of what the stress would be from
a given event and there's also some uncertainty in the
strength of the material, but you're interested in this
little area right here, which would be the area where indeed
the strength of the material is less than the stress that
you put on it, and that area would be an indication of the
failure.
DR. APOSTOLAKIS: You are using the K's there
along this line, right?
SPEAKER: Correct.
DR. APOSTOLAKIS: K is less.
SPEAKER: That's right.
SPEAKER: Not directly, yes.
DR. APOSTOLAKIS: So all this now is the aliatory.
SPEAKER: That's correct.
DR. APOSTOLAKIS: And probably you will put the
epistemic.
SPEAKER: That's correct.
DR. APOSTOLAKIS: I like this figure much better
than figure six in the Maryland paper, although Maryland
tries to go through more detail, but if -- that's what I
mean by coordination. If they could refer to this and then
start developing the algorithm referring to this, that would
be a much better -- by the way, why do you use lambda? Do
you imply a rate?
SPEAKER: These are frequencies of the particular
thermal hydraulic scenario classes.
DR. APOSTOLAKIS: They are rates.
SPEAKER: They are definitely frequencies, yes.
You're ending up with a through-wall crack frequency at the
end.
DR. APOSTOLAKIS: Okay. But the aliatory part
here would be the occurrence of the sequence, something in
the thermal hydraulic, although I don't see how much
aliatory you can have there.
SPEAKER: That's a function of the -- that's
intended -- the primes indicate that you're taking the PRA
event frequencies and then you bend them, so you have a
different frequency, but it's still aliatory.
DR. APOSTOLAKIS: Then on the other side, the way
I understood it is it's primarily the material variability
that contributes to the aliatory part.
SPEAKER: That's where -- again, the variability
is largely the epistemic part, because we're looking at a
specific spot in a specific vessel and looking at the
characteristics of that point there.
The aliatory part comes in the K1C, K1A, and
that's, again, why we need to have this discussion about how
--
DR. APOSTOLAKIS: The variability in K is due to
material, isn't it? That's what it says here.
SPEAKER: That's sort of my gut feeling.
SPEAKER: No. The point is that if you fix -- how
far do we want to go into this, because --
DR. APOSTOLAKIS: We don't have to go into it.
DR. KRESS: It's materials and how you do the
measurement.
SPEAKER: Since George is leaving, maybe you could
spend a few minutes on it. He's not going to be around for
this afternoon's discussion, which is a better place for it.
SPEAKER: The argument in the original white paper
was that even if you knew your material properties
precisely, and it's knowable because you're at a specific
spot in the vessel, you're at the location of the crack tip.
So you could know those properties, but you're uncertain
about that.
And, yes, there are all sorts of uncertainties
that go into your distribution for quantifying that
uncertainty. So it's actually a transformation from the
aliatory uncertainties when you measure to an epistemic when
you're applying it in the calculation. It's all in the
context of the calculation.
But even if you know those properties, some
fraction of the times, your model will predict something and
it will be right, some fraction times will predict something
that will be wrong, basically failure or success of the
vessel.
And it's that fraction that's accounted for by
this P here.
DR. KRESS: K is not a perfect predictor of when
the vessel will fail.
SPEAKER: Exactly. That's the concept we're
trying to bring forward here.
DR. KRESS: In that same context, I know it's
illustrative, but the temperature on the thermal hydraulic
analysis, that's the temperature at the crack location as it
grows, at the tip?
SPEAKER: This is the downcomer temperature.
DR. KRESS: Oh, it's the downcomer.
SPEAKER: It's the environment temperature.
DR. KRESS: I see. You would put that in your
calculation of temperature.
SPEAKER: Exactly. The heat transfer is done.
DR. KRESS: You'd do another calculation.
SPEAKER: That's right, yes. Again, this is just
what the RELAP code will produce, for example.
DR. KRESS: That's the downcomer temperature at
the location you suddenly --
SPEAKER: That's right.
DR. KRESS: -- selected that we're looking at.
SPEAKER: That's right.
MR. WOODS: You'll probably see that one again.
It serves its purpose very nicely. Okay. The status of
where we are.
We are well into the Oconee and Beaver Valley PRA.
We've developed event trees, starting from the IPTS studies.
You remember we did Oconee before and we didn't do Beaver
Valley before, but H.B. Robinson is similar enough, so you
can start with those event trees.
We're using generic initiating event frequencies
and top event split fractions from industry data to focus
and develop and decide where to work on the model.
We are developing the fault trees for Oconee,
where you have data for the top events. In other words,
instead of just putting in a feedwater system fails, if you
have enough data to support what part of it failed, then you
would want to develop a fault tree to use that data.
We are putting in potential human failure events
developed from the Athena team and the quantification of
these things is currently ongoing.
We could give you more details, but I'll try to
leave it with that.
The other two things we intend to do are to review
the analyses that are done by the licensee for Palisades and
Calvert Cliffs and at the moment, what we're doing, we've
collected a great deal of information from Palisades and
some information from Calvert Cliffs, reason being we're
going to do Palisades next after Beaver Valley, and we are
assessing basically the adequacy of the information, but we
really haven't reviewed, started the detailed review of
those plants.
Now, before I get to the next slide, I want to
tell you, please, you're not supposed to try to read this.
I do have a magnifying glass in my pocket that we might have
to use to read it.
But the objective of showing this slide is to show
you that -- I think I can stand up here. This is part of an
event tree. It's not even the whole event tree. This is
the event tree for the initiating event and reactor trip,
and this is reactor trip and it trips.
Then across the top we have all the different
things that can happen or not happen and you probably can't
even read that, but the point is -- one point I want to make
here is we are developing, in further detail, a different
part of the tree from what you're used to probably, because
we're worried about pressurized thermal shock, which is an
over-cooling event.
Usually, when you do one of these event trees,
you're worried about core damage directly from failing to
provide cooling.
So you tend to develop the top side of the event
tree, where you have what normally would be successors, like
the HPI comes on, but it stays on and normally that would be
fine, but you need to develop that further to analyze the
over-cooling.
And what I'm going to do with the next three
slides, I think it is, is show you the details of this
slightly darkened path, if I can follow it. It goes on over
here and ends up on 14 or 15, whichever one we decided, but
to kind of walk you through that.
SPEAKER: Now, just from your comment there on
success in the normal PRA, are you arguing that many
conventional PRAs then don't pay enough attention to the PTS
event?
MR. WOODS: Conventional PRAs may not even include
risk from PTS at all.
SPEAKER: Okay. Because you're assuming it's
screened out.
SPEAKER: Because embrittlement is not an issue.
SPEAKER: Yes. As long as you're not embrittled,
who cares.
MR. WOODS: It's a good point, but they're not
there at all. They're not there at all.
SPEAKER: So you really have to develop these
event trees yourself.
MR. WOODS: Yes.
SPEAKER: You can't get them from the plant PRA.
MR. WOODS: That's the point.
DR. APOSTOLAKIS: They start with the plant.
SPEAKER: They start with it.
MR. WOODS: That one was developed from -- I just
took it down, but --
SPEAKER: If I could comment. We certainly use
the plant PRAs to the extent we can, but a lot of the
information comes from the earlier IPTS studies, which did
develop event trees, and we've expanded on those and
customized them for the studies we're doing.
MR. WOODS: I wanted to mention this, 163 end
states, you can't read that number, but that's what it says.
And this is only part of this one tree, because this
particular thing is turbine bypass valves sticking open.
This is none, one, two, and four.
So these two lines would lead to equal size -- each of them
would lead to an equal size of what's shown there and then
this has to do with the PORV or the primary side safety
valves sticking open, one or the other. So here's two more
lines that would lead to something that looks like that.
So that's a fourth or less of that one event tree.
DR. APOSTOLAKIS: So this is not a binary tree
anymore.
SPEAKER: Correct. That's correct.
MR. WOODS: And that's one of like six to eight
event trees, depending on which plant you're talking about.
There's one for steam line break and LOCA and whatever, in
addition to this tree.
DR. APOSTOLAKIS: So where in the tree do you have
human actions that ATHENA will come in to help?
MR. WOODS: That's coming up. That's why we
wanted -- one of the reasons we wanted to work through one
of these.
Now, this is the part that I showed you that was
highlighted and the next two slides have the words, some of
the words that I intend to use describing this. So I'm
really using three slides at once here.
But walking through this, I think I can point
better if I stand up. Okay. You start with a trip and the
first question is does a PORV or a safety valve on the
primary side stick open, and the one that we've chosen to
use as an example, we say that it doesn't. It's okay.
So having decided that it doesn't -- I mean, it
doesn't open; so, therefore, it can't stick open, so it just
goes straight through here.
But we do say that one turbine bypass valve sticks
open. You know, the turbine bypass valve would stock open
on a trip, in a turbine trip, because you've got to dump the
heat somewhere. So it's supposed to open, but it's not
supposed to stay open.
So we say that one sticks open and the operator
doesn't isolate it. Now, there's the first human event that
you have to look at.
And in this particular case, the ATHENA team
returns a table to the PRA analyst that says, okay, here's
the probability that he won't isolate given that there's no
other complicating factors in the plant or given that
something else is going on that might distract him or
mislead him or whatever, or several things maybe.
You might have two or three different numbers,
depending on the -- you choose which one you use depending
on the circumstances.
So this would be the one where it's most probable
to isolate it, because nothing else is going on in the plant
yet.
And then the main feedwater in Oconee, this is the
Oconee scenario, the main feedwater is supposed to run back.
That's the normal situation for this event.
But in this particular example fault tree, we say
that instead it trips and then the emergency feedwater comes
on and the normal situation that would control to a certain
level, but instead it over-feeds both steam generators. The
others are also in this event tree. I'm just showing you
the example of the one where it over-feeds both steam
generators.
And I guess there's a failure to recover. I
missed one.
SPEAKER: No, it doesn't matter. It doesn't
matter, because that's a fail to start.
MR. WOODS: All right. So that's the secondary
side. On the primary side, because of the over-cooling, the
pressure goes down and the HPI comes on at 15 to 1,600
pounds. Anyway, it goes low enough so it comes on.
And so we follow that part of the tree and the
last part would be whether or not you lose subcooling and
the main reactor coolant pumps trip or they don't. In this
case, we don't think we would lose subcooling, but we're not
absolutely sure of it.
So there's another split here that says it trips
or it doesn't. Then, finally, there's another split here,
which is another human factor, where HPI flow is throttled
or it's not, and then in this case, you don't take the
simplest no load type human factor. You take the one where
other things have already gone on, because you've already
had a stuck-open turbine bypass valve and you've already
failed to control the emergency feedwater.
So other things are going on in the plant and
it's, therefore, less likely that he will remember to
throttle the HPI, and they use a different number.
DR. APOSTOLAKIS: So ATHENA now has a way of
telling us how likely it is.
MR. WOODS: Yes. It's not exact, of course, but
based on their experience and the data that they've seen and
the simulator runs that they've seen for that sort of thing,
they do have an organized process by which they come up with
a table. But it probably has a name.
SPEAKER: No. This is just -- basically what
they're doing is a self-elicitation of the group. The group
discusses the event. The probabilities are chosen on a very
coarse scale. It's one of four values, it's either .5, .1,
.01 or .001.
So basically you're corresponding to notions of
likelihood given the scenario. There is no attempt to make
it any finer than that. And the group discusses it, brings
up the reasons why the failure might occur, what sorts of
things might prompt a failure, and then says, well, given
the circumstances, given our observation of the operating
crew, that performance scenario is for us, given our
understanding of the procedures, talking with the training
supervisors, here is what it is.
MR. WOODS: Okay. The explanation, like I say, is
on the next two slides. I hope that what I said is what's
on the next two slides. I'm not going to go through it now
and make sure I didn't miss anything. You can look at it
later. I think it's self-explanatory, or largely so.
I'll go on to slide nine. Information used in the
analysis. The point of this slide really is just to show
you that we don't just take a cursory look at these plants.
We collect quite a bit of information and it's all listed
there, and I don't think there's any need, again, we're
running very late, to read that to you.
But we basically start with the IPE and sections
of the FSAR and the P&IDs that are available. We collect
all the emergency operating procedures, some of the abnormal
operating procedures, because they give you an idea of human
actions that lead to the PTS initiators.
Then down about a little over halfway, training
provided to the operators is something that we really
concentrate on. The ATHENA team has actually witnessed a
simulator practice in both of the plants, in Oconee and in
Beaver Valley. They asked for operating experience from the
two plants on very related and relevant like PRVs, SRVs,
whether they stick and that sort of thing.
Now I've lost my next slide. I got my files mixed
up, sorry. We've probably discussed a lot of that.
Obviously, we're using the better operating experience. We
got three or four times more operating experience than in
1980 when we did this before, and, also, that will
contribute to the initiating event frequencies and also to
the failure probabilities.
We are using current plant design and operating
procedures. Some of the procedures are even new
specifically to avoid this kind of event since 1980, and
that makes a big difference.
We think we've got better coupling between the
event sequences and the TH, because we've got capability to
run more TH scenarios. Things are on a PC now instead of a
$100,000 per run for a RELAP run on a big machine.
I'll go on to the next one. It's a continuation
of this one. I think some of the main things are on this
slide, actually. We do think, and it's already come up,
that we are taking contextual factors affecting the operator
into account much better. In fact, I'm not sure it was even
done at all back in 1980.
I mentioned that there's two or three different
numbers that you choose from based on whatever is going on
in the plant other than that particular event. We're doing
that.
The last two bullets really are meant to show that
we are using this to take into account the pluses and the
minuses. With the new human methods that we have, we are
better able to take into account errors of commission. In
fact, we're able to try to take them into account. We
didn't even attempt before.
Such things as operator trips, RCPs when he's not
required to do so, or the operator isolates the wrong steam
generator or whatever. When that comes up in a tree like
this that we have a number to put in, which is not an exact
number, but it's better than no number, we think.
Also, on the other side of the coin, then, like
when we went to one plant, we could see that they were
trained on not having safety injection on when you didn't
need it on. It's one of the first steps that they go to and
one of the procedures they always go to and they drill on it
and we just -- the ATHENA team just thinks that it's very
unlikely that they'll forget to take that action.
So previously, where we might have had a fairly
high probability of that, it isn't anymore. So it's a
balance and it's a representation of the plant more as it
really is rather than as you might think it would be from an
analyst bench.
Concluding remarks. There's not much new to say
here either. We think we are able to screen out some event
sequences that won't be a problem and like normal trips and
the vents that don't cool down past a certain point, and we
don't spend a lot of time, waste a lot of times on events
that won't be a PTS problem.
We think we're doing a better characterization of
the event sequences, better binning of them, especially for
Oconee. That's not to say we aren't doing a good job on
Oconee and Beaver, it's just to say that they didn't do a
very good job on binning things in Oconee back in 1980.
They dumped most of the things into the "other" category and
then ended up giving it a much higher consequence than they
should have. So we are certainly correcting that.
We mentioned we think we're doing an improved
treatment of uncertainties, which Dr. Apostolakis wants to
hear more about and we will do that.
The issues are we haven't yet handled external
events like we want to. An example of that would be a fire.
A fire could certainly burn up some cables and cause all
sorts of problems at once that maybe we haven't taken into
account by the analyses that we've already shown you.
We do know that when we get through with the four
plants, we will have two analyses that we've done, being
Oconee and Beaver Valley, and we will have two analyses that
we have reviewed, that will be Palisades and Calvert Cliffs,
and there are bound to be inconsistencies and we're going to
have to come to grips with how we handle --
END TAPE 1, SIDE 1.
TAPE 1, SIDE 2 FOLLOWS:. BEGIN TAPE 1, SIDE 2:
-- four analyses that really are on a different
basis. You're kind of trying to put apples and oranges in
the same bin and we have to deal with that.
Then the generalization is that obviously we've
got four analyses and we're going to have to try to use that
to represent the risk at all the plants, and we have some
idea how we're going to proceed to do that.
Then the acceptance criteria, which we mentioned,
is sort of a separate presentation.
That's all I had. And I'm sure we're way over,
but we'll answer any questions that we can.
SPEAKER: Thank you.
SPEAKER: A non-controversial one.
SPEAKER: Sure it is.
MR. BISSETTE: My intention was just to give a
brief summary of where we stand on the thermal hydraulics
part of this three-part program, just so you have all the
pieces.
I'm David Bissette, from the Thermal Hydraulics
Branch in Research. The objective of the thermal hydraulics
work is to ensure that for the risk-significant classes of
events, the thermal hydraulic input developed at the time of
the IPTS study back in the early '80s are still operative or
updated as needed, provided the uncertainty, estimating
uncertainty of these calculated values, and, as you heard
before, the IPTS study, there were three PWRs selected for
analysis, one from each vendor, Oconee, Calvert Cliffs and
H.B. Robinson.
And as you've heard, in the current study, we've
switched to a fairly similar, also a three-loop plant,
that's Beaver Valley.
These are the principal thermal hydraulic issues
that we encounter in single and two-phase loop natural
circulation, criteria for interruption of loop flow which
causes flow stagnation in the cold leg and downcomer, number
of cold legs which are supposed to be flowing to assure
mixing in the downcomer, local fluid, fluid mixing, and
non-thermal stratification in the cold leg, plume, this is
the plume entering the downcomer, plume mixing in the
downcomer, and all these are being studied in the
experimental program underway in the APEX facility.
DR. APOSTOLAKIS: Are you going to do a detailed
uncertainty analysis, just as the other guys are proposing,
the fracture mechanics? I mean, you're going to identify
model uncertainty and parameter uncertainty and everything,
or thermal hydraulics is immune to that.
DR. KRESS: There is no aliatory.
DR. APOSTOLAKIS: I know.
MR. BISSETTE: It follows along similar lines.
DR. APOSTOLAKIS: Flaws of nature.
MR. BISSETTE: It's being done also at the
University of Maryland. Do you want me to say more about
it? Do you want to say anything?
It's kind of a combination. Well, in the thermal
hydraulics area, the way we treated uncertainty to this
point in time is sort of the CSAU methodology, which you
probably all have some familiarity with.
What that is is you identify the most important
phenomena and you -- for each phenomena, you find the models
in the code that models phenomena and you vary them
according them according to the uncertainty, which you
physically understand the phenomena.
So you run the code repeatedly and you see the
sensitivity on the final answer that you're interested in;
in this case, it's pressure and temperature in the
downcomer.
SPEAKER: Dave, maybe I can -- the short answer is
yes, we're trying. We're taking our best shot. Some issues
we think we can handle reasonably well, like what happens in
the scenarios where it's basically single phase. For
two-phase scenarios, it's more complicated.
That's certainly where the model uncertainty
issues arise. For the single phase kinds of situations, it
looks more like it's an input parameter based on what's
happening in the event sequence, which has only been defined
to a certain level of detail.
So when exactly is a particular action taken, for
example, that's an aliatory issue which we need to reflect
in the results.
We're in the process of still developing the
methodology and we're test applying it to Oconee. There's
been ongoing discussions among the PRA thermal hydraulics
and thermal hydraulic uncertainty analysis groups, but I
don't know that we have a great answer for you at this
point.
Again, it sounds like something that would be
worth talking about in the meeting when we talk about how we
deal with uncertainty.
DR. APOSTOLAKIS: But the amount of effort will be
the same.
SPEAKER: It's a significant effort on our part, I
think.
MR. BISSETTE: The plant, the first plant we
started with parallels the other efforts, it's Oconee.
We've been performing analyses using RELAP. Thus far, we've
calculated 25 transients with RELAP-5, Mod 3. I don't know
if you recall, but the picture Roy Woods showed, basically,
these are 25 transients out of the hundreds of thousands of
the sequences that he showed on his event tree.
We've run these - the objective was to run these
transients to at least 10,000 seconds. We have achieved
that. This is a significant improvement over the former
study, where most of the transients were only run out to
about one or two thousand seconds and extrapolated out to
two hours. Just some fairly simple straight line
extrapolation.
Also, contrary to the earlier study, we modeled
the downcomer as a two-dimensional configuration as opposed
to the former 1-D that was used before.
SPEAKER: In the past, with RELAP, there have been
some problems with running out for extended periods of time.
Was this stable?
MR. BISSETTE: It was surprisingly stable. We had
a few code failures, but we were able to run through them by
reducing the time step. So for all the 25 cases, we went
out to 10,000 seconds.
So I found it remarkably stable compared to what
sometimes we've experienced in the past.
DR. KRESS: You don't use REMIX at all anymore.
MR. BISSETTE: We are going to use REMIX. We are
using REMIX and I'll mention that a little bit further on.
I don't have too much to say about it.
So this is going to show you the conclusions from
the 25 cases that we've generated so far. Rather a useful
interchange between the PRA and the thermal hydraulics work,
and I think that's also an improvement over the old IPTS
study in the early 1980s.
We've covered a great spectrum as part of the work
we did for Oconee and it covers a range of interests, but
the primary system pressure phase, let's say, high near the
secondary side pressure, to where the primary system
pressure drops below the accumulator pressure and further
down to about 200 psi, which is the low pressure injection.
Finally, results are sensitive to the trip
criteria for the reactor coolant pumps. Procedures call for
tripping the reactor coolant pumps on loss of subcooling.
And once subcooling is lost in a small break LOCA,
it will generally not be reestablished unless the break can
be isolated. So that means that when the pumps trip, they
stay off.
We find that we've done combinations of primary
side and secondary side failures. We find that when we
combine secondary side failures, like stuck-open valves
with, say, a small break on the primary side, it helps
maintain subcooling and, therefore, reactor coolant pumps
are not tripped.
Like I say, it's a big difference if you trip the
pumps or not, because if the pumps are running, basically
you have a tightly coupled system between all the loops in
the primary side and all the generators on the secondary
side. So your heat sink is very large compared to a
situation where you trip the pumps and now your focus is
just on the volumes of water associated with the downcomer.
So when you trip the pumps, at least for Oconee,
stagnation begins very quickly. The downcomer cools in
response to the high pressure injection. And, finally,
comparing the primary side between breaks in the hot side
and the cold side for a given break size, hot leg breaks are
a little bit worse than the cold leg breaks. That just
confirms something we saw in the old IPTS study.
We have a small activity right no to coupling
REMIX with the TRAC code. We're also going to run -- so
we've got a couple the REMIX and TRAC code and run a
two-inch break with the coupled code.
We're also running the same break with REMIX using
the boundary conditions that come out of the RELAP
calculation.
So this is right now what we're doing with REMIX
in terms of the calculations.
SPEAKER: So it's not a thermal hydraulic type.
What is this giving you?
MR. BISSETTE: It gives you -- REMIX gives you
another indication of downcomer temperatures.
SPEAKER: REMIX is a two-dimensional code.
MR. BISSETTE: REMIX is, let's say, basically a --
it treats the mixing volument of interest. REMIX applies to
stagnant flow conditions. It treats a part of the system
that's of interest, which is the cold leg, the HPI
injection, the downcomer and the lower plenum.
It treats that as, let's say, a single volume that
has five mixing regions in it. And the mixing regions are
treated on the physical basis of -- based on like through
number treatment of mixing and stratification and plume
dissipation.
So it's basically a physically based engineering
tool to give you mixed temperatures.
DR. KRESS: That's stuff you can't get out of
RELAP.
SPEAKER: Yes, right.
DR. KRESS: And you really need that.
SPEAKER: So it basically comes in when you get
the stagnation.
MR. BISSETTE: That's right. You use some
boundary conditions that you get from RELAP, depicts the
inlet and outlet boundary conditions.
Our plan is to repeat selected cases that we've
done already with RELAP and we're just about to get these
calculations underway and we'll have the results in about
one month.
Now, this is the only further slide I had on
uncertainty evaluation. This is the study that's being
performed by the University of Maryland for Oconee. I had
mentioned CSCU before. What they are also doing is we came
up with a simplified model of the Oconee system. It was
based on simply conservation of mass of energy, and
performed calculations.
DR. APOSTOLAKIS: Model uncertainty there.
MR. BISSETTE: So my advice is we --
DR. KRESS: We don't have a write-up on that, at
least I haven't seen it.
DR. APOSTOLAKIS: I haven't seen it either. Is
there anything we can read about it?
MR. BISSETTE: We have a partial draft report
that's in preparation. There should be, let's say, a first
draft in a few months. There's nothing really -- right now,
it's not much more beyond viewgraphs you can look at.
DR. APOSTOLAKIS: That should be part of whatever
subcommittee meeting.
DR. KRESS: Yes. This should be part of the same
one, when we talk about the other one.
MR. BISSETTE: Yes, it should be. I had mentioned
the testing program we have underway at APEX. APEX is
located at Oregon State University. The objective is to
provide experimental data on the thermal hydraulic PTS
issues, as well as for code assessment. We did a scaling
evaluation to compare the APEX facility to Palisades and, as
far as that goes, to other CE plants, like Calvert Cliffs
and Fort Calhoun.
APEX was originally configured to model AP-600.
CE plants are similar in size to AP-600. The facility is
modified to add loop seals, HPI connections to the cold
legs, and additional thermocouples in the cold legs and
downcomer.
We performed pre-test calculations using RELAP and
REMIX. There's REMIX again. We conducted our first test in
August and the remainder of the test program is scheduled to
be done by the end of the calendar year.
I'm just going to show you -- put this up just to
briefly show you what the facility looks like. It's a
two-by-four arrangement, similar to the CE plants, with two
hot legs, two generators, four reactor coolant pumps feeding
into four cold legs.
Then this is just a top view, comparing the APEX
loop layout with Palisades.
DR. KRESS: The injections in the hot leg are all
the same. High pressure injection is in the hot leg?
MR. BISSETTE: No, the cold leg. Because all
plants have connections to the hot leg, as well as the cold
legs for the injection systems, but normal injection path is
the cold legs.
So I won't go through the test matrix, in the
interest of time, but here it is. You can look at it.
Basically, the tests are PTS sequences, in addition to more
basic and separate effects kind of testing to cover the
range of issues that I had mentioned earlier.
Now, in addition to Oconee, we will be doing
Beaver Valley, Calvert Cliffs and Palisades. We haven't
done any calculations thus far beyond exercise the input
models for these plants. We started converting H.B.
Robinson decks to Beaver Valley.
We're scheduled to have a set of calculations
completed by January of the coming year and we'll follow
that with Calvert Cliffs and Palisades, hoping to have the
calculations by March of next year.
The final slide is we have our Oconee
calculations, RELAP, ready for transmittal to Oak Ridge.
They use them as boundary conditions for FAVOR. We expect
to provide the calculations for a Westinghouse three-loop
plant based on Beaver Valley by, say, early the coming year,
and Calvert Cliffs and Palisades by the middle of next year.
SPEAKER: Are the thermal hydraulic boundary
conditions for Oconee different than they were in the '81?
I mean, have they changed substantially?
MR. BISSETTE: I haven't done a -- we haven't
looked at that in detail yet. That's something that we will
be doing in the next month or two.
DR. KRESS: Well, you had a single curve for the
pressure and temperature.
MR. BISSETTE: Yes.
DR. KRESS: But now you're going to have a
distribution.
MR. BISSETTE: Well, we may have a curve with that
uncertainty band on it.
DR. KRESS: There may not be much uncertainty
about the pressure, but there can be about the temperature,
I guess.
MR. BISSETTE: Yes. What we've found, in terms of
looking at the phenomena, is that a lot of the phenomena are
pretty well -- we believe the dominant phenomena are pretty
well modeled by RELAP. There are some uncertainties because
you can't model two fluids, two liquids in a one-dimensional
code.
DR. KRESS: That's your cold liquid and your hot
liquid.
SPEAKER: Which is one of the reasons we use
REMIX, too. What we're going to do, we're going to hold off
on the discussion of the probabilistic fracture mechanics
until this afternoon and we'll have all the probabilistic
fracture mechanics discussion together.
We're going to take a break now and then come back
and go into the flaw distribution discussion.
SPEAKER: Great.
SPEAKER: So be back at 10:15.
[Recess.]
END TAPE 1, SIDE 2.
TAPE 2, SIDE 1 FOLLOWS:. BEGIN TAPE 2, SIDE 1:
SPEAKER: The next discussion is the generalized
flaw distributions, and I guess that's Debbie Jackson and
Lee Abramson will be making the presentation.
The first test is in. Okay. Passed.
SPEAKER: They found it.
MS. JACKSON: I'm Debbie Jackson, and Lee
Abramson. We're going to present the results from the
expert judgment process for the development of the flaw
distribution.
The first two slides just go over a little bit of
background information and reasons why we are doing this
flaw distribution. The last major work on flaw distribution
was done in the mid '70s and early '80s. It was a Marshall
distribution, and that was done not only with nuclear
vessels, but also with non-nuclear vessels. So this work
that we're doing now is a lot more expensive than the
previous work on the Marshall distribution.
This slide just discusses a few of the reasons why
we decided to do an expert judgment process for development
of the generalized flaw distribution.
This is a list of the fabricators for domestic
reactor vessels and the list is in order of the percentage
of vessels that were manufactured by each organization.
The last Rotterdam and Society Crusoe, they
finished the fabrication. One of the fabricators, Babcock
and Wilcox, ran behind schedule during their fabrication
processes. So some of their vessels were finished by the
Rotterdam and Society Crusoe.
This is a slide that lists the reactor vessel
material that's been inspected by PNNL that's going to --
that was used for the flaw distribution. The Midland vessel
was inspected in the early '80s and it was with a different
type of SAF-UT system, and since the Midland inspection, the
UT exams have advanced a lot.
So the inspection techniques were different, so
we're actually not going to include the Midland data. We're
only going to do the PVRUF-C, Shoreham, the River Bend and
Hope Creek vessels.
SPEAKER: PVRUF, what is that?
MS. JACKSON: Pressure vessel research users
facility. It's a cancelled vessel that was at Oak Ridge,
and we've used that.
SPEAKER: Debbie, I couldn't find it anywhere in
the report. The three boilers, Shoreham, River Bend and
Hope Creek, what are the weld processes that are used there?
MS. JACKSON: The weld processes for those were
submerged arc and then they are back-gouged with -- the
inner sods were done with submerged arc.
SPEAKER: How about the axial welds?
MS. JACKSON: The axial welds were -- I believe
they were submerged arc, but some of them may have been
electroslag. I need to look that up.
SPEAKER: Okay. I just wondered if we were mixing
electroslag data in with the other data.
MS. JACKSON: Not with the -- not for the PTS, no.
We don't have a lot of data on the electroslag weld
processes, because a lot of that was done with the boilers.
SPEAKER: Okay. But I just wanted to make sure it
wasn't being included for the PTS study.
MS. JACKSON: During the examinations that PNNL
was doing, we came up with categories to categorize the
different flaws and what we came up with were different
regions of the vessel.
The inner region is the NR-25 millimeter, the
inner one inch. The outer region was the outer one inch,
outer 25 millimeter, and the mid region was the remaining
part of the vessel wall.
Volumetric and planar, we have the weld, the clad
and the base metal and repair weld versus non-repair welds.
We found out from some of the data that there are
quite a few flaws in the repaired areas of the vessel.
These next two slides are going to go over just
the steps that we used in the expert judgment process. We
first defined some of the issues, determined the level of
complexity. We identified an expert panel. We sent some
issues to the panel.
The panel had a meeting and we had elicitation
training, which was performed in Atlanta by Lee Abramson.
DR. APOSTOLAKIS: I have a question here.
MS. JACKSON: Yes.
DR. APOSTOLAKIS: I think I read the report and,
in my opinion, it's not clear how the expert judgment was
used, what the objective of the elicitation was, and I
formed an opinion after I read the whole report, and please
correct me. This is my impression.
That you actually started with a distribution for
the size, the crack depth, and also for the density that is
based on data and what you did with the experts is you
modified that, depending on the various things that you have
here, on whether it's unrepaired weld metal or unrepaired
cladding or the various other things that you have here,
plate versus welds.
Is that correct? In other words, you did not
elicit from the experts information that would give you the
actual density. You didn't ask them that.
SPEAKER: That's correct.
MS. JACKSON: No, we did not.
SPEAKER: That's correct.
MS. JACKSON: You're right.
SPEAKER: We just asked relative values.
DR. APOSTOLAKIS: Relative values.
SPEAKER: That's correct.
DR. APOSTOLAKIS: So I suggest, since this is
still draft, that you add a section someplace explaining
this, because I was really trying very hard to understand
what was going on and then you hit me on page 25 with all
the information that comes from statistics and then I had to
figure it out myself.
MS. JACKSON: Okay. That's a point well taken and
we'll make those --
DR. APOSTOLAKIS: And also it would be useful if
you showed how these various factors were used, what was the
arithmetic, in other words.
SPEAKER: Okay. There is a considerable -- there
is some detail in the report as to how --
SPEAKER: It's sort of lost in those notes.
SPEAKER: Well, it's in the notes.
DR. APOSTOLAKIS: The report says that this is the
mid value of the median and so on.
SPEAKER: Correct.
DR. APOSTOLAKIS: And using that, we get. And I
guess the "using" is the thing, how exactly -- I mean, maybe
it's a simple multiplication.
SPEAKER: It is, yes.
MS. JACKSON: The report has been revised since
then and the first revision, because it's going to be
revised quite a bit before the final NUREG comes out at the
end of next year, but the notes have been revised
extensively.
SPEAKER: I'm not sure which version you saw,
George. The second version hopefully will be more explicit
and the intention of the notes was to give you a road map to
let you reproduce the calculations yourself without a great
deal of trouble.
SPEAKER: But George is right. You really ought
to separate the ones where you're working from data from the
ones where you've essentially modified the distributions
based on the --
SPEAKER: This is -- we tried to make this
extremely explicit in the notes.
DR. APOSTOLAKIS: Yes, I understand that.
SPEAKER: And some of the numbers that we got --
SPEAKER: Well, I ended up highlighting my table
so I could tell which was which.
SPEAKER: That's right.
MS. JACKSON: That's been revised.
DR. APOSTOLAKIS: I've read already 26 pages. On
the 27th, there is note number three, these values are
multiplied by -- this is such a big thing, it should be up
front some place that this is what the objective was. We
will rely on statistical data to get density and --
SPEAKER: We'll try to make it more explicit.
DR. APOSTOLAKIS: -- distribution. And the reason
why we have to go to experts is because the data is a
mixture and you can't tell where it comes from, because if
you could, then you wouldn't need the experts.
Then you had to modify it or to adapt it to the
particular circumstances of interest.
SPEAKER: That's right.
DR. APOSTOLAKIS: And that's what we're doing,
that's what we're eliciting. Okay. And then here, and, for
example, for this factor and this factor and this factor, to
get this, we multiply this by that. That would go a long
way towards helping the reader really place it in context.
SPEAKER: Okay.
MS. JACKSON: Okay.
DR. APOSTOLAKIS: Good.
MS. JACKSON: Thank you for that.
SPEAKER: Just on that, too, I mean, you give the
tables up front for the distribution and the PVRUF and I
can't make the numbers add up to get the numbers in table
5-1 for the small flaws and large flaws and greater than
five millimeter flaws, and you're referring me back to the
original PNNL report.
You ought to just bring those tables from the PNNL
report and put them in here so that --
DR. APOSTOLAKIS: And I would like, by the way, to
get your reference ten, Shuster, Dr. Hessler,
characterization of flaws in U.S. reactor pressure vessels.
It's a NUREG published in 1999.
It seems to be an important document in this
context.
MS. JACKSON: It is. There's three --
DR. APOSTOLAKIS: So if you can send me a copy, I
will appreciate that.
MS. JACKSON: We have copies.
DR. APOSTOLAKIS: NUREG-CR-6471.
MS. JACKSON: There's three volumes of that now.
The third volume just came out.
DR. APOSTOLAKIS: That will do it. Three volumes.
I would like to get that.
SPEAKER: Beginning to get indigestion, George.
DR. APOSTOLAKIS: That will teach me.
SPEAKER: There's 10,000 flaws, George. When you
discuss each one --
DR. APOSTOLAKIS: Each one, what happened.
MS. JACKSON: Okay. Well, that will be good.
This next --
DR. APOSTOLAKIS: Now, just to -- you know, we
have to be nitpicky here. How the hell do you know it was
successful? You just got some numbers and you used them.
Why was it successful?
MS. JACKSON: Because we completed 17 --
MR. HACKETT: This is Ed Hackett. I think I could
speak for Debbie and Lee, because they're going to be humble
and modest. But I think it's just the fact they made it
through and people didn't die in the process. So it was
kind of -- maybe this is a low bar for success, but at least
that was part of it.
MS. JACKSON: Our first elicitation session was
with Vic Chapman. He's one of the authors of the Marshall
report.
DR. APOSTOLAKIS: I know him.
MS. JACKSON: And the session lasted --
DR. APOSTOLAKIS: You don't call him Lord
Marshall?
MS. JACKSON: Retiree Marshall, now, Retiree
Chapman.
DR. APOSTOLAKIS: He's bored.
MS. JACKSON: But this session was borderline nine
hours. So after that, we decided we had to make some
changes.
And I say it's evolving because the first few
elicitation sessions that we did were different than the
final few. Each session, we learned some additional
information from the experts. One thing in particular, we
had cladding as a group in itself and then one of the
experts suggested that we break cladding down into the
different specific methods of cladding, strip cladding,
multi-wire and single-wire.
With that, we had to re-elicit the experts after
we finished the final elicitation session, because there
were so many changes throughout the process.
DR. APOSTOLAKIS: Are you eliciting the experts or
their opinion?
MS. JACKSON: We elicited the experts to get their
expert judgment and opinions on some things.
This is a list of the areas of expertise we had
for the different experts.
DR. APOSTOLAKIS: Now, I have another question
that's not on the viewgraphs. You say here in the report
that in addition to the empirical data, PNNL has used the
flaw simulation model of R.R. Prodigal to estimate the
numbers and sizes of flaws in the welds of the PVRUF and
Shoreham vessels.
To estimate the number and sizes. What kind of a
code is that? What input do you put in there?
DR. KRESS: That's an expert.
DR. APOSTOLAKIS: It's another expert.
MS. JACKSON: It's an expert.
DR. KRESS: It's expert-based code.
MS. JACKSON: Prodigal was done some years ago and
it was another expert judgment, as you said.
SPEAKER: It puts a flaw in and then has a
probability distribution for whether that flaw then goes to
the next bead in the weld, depending on what you're doing.
MS. JACKSON: It simulates a weld, the given
welding process.
DR. APOSTOLAKIS: This is a different use of
expert judgment. Now you're referring to density. I would
like to have that, too.
MS. JACKSON: Okay. And that was one of the
comments we got. We need to provide some additional
explanation on the Prodigal code in that report.
DR. APOSTOLAKIS: The commitment by the NRC's
Office of Research to develop a generic flaw distribution
has been received positively by the NRC's Advisory Committee
on Reactor Safeguards. We said that?
DR. KRESS: Yes, we said it was a good idea.
SPEAKER: With the Marshall flaw distribution.
DR. KRESS: Yes.
SPEAKER: Too long.
SPEAKER: Yes.
DR. KRESS: Yes.
SPEAKER: Even if he's a Lord.
DR. KRESS: In fact, I think we said if you could
do that better, you could go a long way to solving the whole
problem of PTS.
DR. APOSTOLAKIS: A lot of questions have a depth
for information.
DR. KRESS: I see. You've got too many answers.
MS. JACKSON: The next two slides have three
definitions that were developed for the flaw distribution
for this process, and for consistency, we developed a
definition for the flaw. This was done through a consensus
process with the experts and the definition is an
unintentional discontinuity that has the potential to
compromise the reactor vessel integrity and is in the vessel
after pre-service inspection.
[Tape stopped and restarted.]
MS. JACKSON: We began to use the definition that
was in ASME and some of the experts felt that was
inappropriate. So this is what we came up.
DR. KRESS: So if it's an intentional one, it
doesn't count.
MS. JACKSON: Right. If it's an intentional -- if
the base metal dinged during travel or something like that.
And two additional definitions were for a small flaw and a
large flaw and that's additionally broken down into a small
flaw in the weld metal and cladding and flaws in the base
metal.
We developed a list of --
SPEAKER: When you do that, it would be helpful if
you gave us bead sizes then for each of the welds we're
looking at.
MS. JACKSON: Yes, because the bead size does vary
so much with the different processes.
SPEAKER: I couldn't back that out of the reports.
SPEAKER: The bead size range, I think, is in the
tables, in one of the tables, 5.1.
MS. JACKSON: Or some of them.
SPEAKER: Or some of them. We gave the range of
bead sizes in there. It varied.
SPEAKER: Everything is in 5.1, if you can find
it.
DR. APOSTOLAKIS: How come you don't name the
experts?
MS. JACKSON: We do have them now in the backup
slides. We've listed --
DR. APOSTOLAKIS: It's here?
MS. JACKSON: Yes. That was one of the difficult
processes, because many of the people who were actually in
reactor vessel fabrication are retired and some of them are
no longer here. So that was kind of a torturous process.
I almost called someone and then someone informed
me that the person had just passed away. So I didn't make
that phone call.
SPEAKER: That's a hard call.
MS. JACKSON: This is the list of issues. We
tried to come up with a comprehensive list so that we would
include every aspect of reactor vessel fabrication and all
of the different areas where a potential flaw could be
introduced during the fabrication process.
DR. APOSTOLAKIS: This is now another interesting
point here. Since you're planning to adopt distribution
that's based on data using information from these elements,
is there a possibility that you are considering too many
issues and that may lead to too many factors multiplying
things?
In other words, you are going to such detail that
you may start getting optimistic results. And were the
experts asked?
MS. JACKSON: Yes. I'm going to go --
DR. APOSTOLAKIS: And, also, I'm not sure you can
treat these things as independent.
MS. JACKSON: That was one of the things
throughout as we learned through the process. We broke the
characteristics down. Some of the characteristics, the
experts were able to give us quantitative numbers.
I'm going to explain how we got information from
them regarding the introduction of a flaw, but in the end,
we found out that most of these in this column and some in
this column -- oh, I'm sorry.
DR. APOSTOLAKIS: See, my point is it's the same
like in a fault tree. You can go way down into detail and
--
DR. KRESS: But in this case, it's like entropy,
though. It just broadens the distribution, the more you put
in it.
DR. APOSTOLAKIS: No.
DR. KRESS: I think it does.
DR. APOSTOLAKIS: Because they multiply by
fractions the various -- the statistical density.
DR. KRESS: That broadens it, though.
DR. APOSTOLAKIS: They haven't done any
uncertainty yet.
SPEAKER: They blur the resolution, but it should
keep --
DR. APOSTOLAKIS: Keep it down, because now you
have -- and because of filled versus short and then welder
skill are multiplied independently.
SPEAKER: As Debbie said, some of these are
qualitative and some are quantitative. It's only the
quantitative and actually in the left-hand column --
actually, more than half are qualitative, and I'll explain
more in detail when I give my presentation.
And when you look at the report, we used data
wherever possible and there was quite a bit of data. We
only used the expert judgment to fill in when there wasn't
any data.
DR. APOSTOLAKIS: Yes, but that was not really my
question, because if you have a bunch of experts and you
give them the issues, then they tend to focus on, okay, what
does product four mean, is it important and so on.
But if you look at the whole list, are these
really independent characteristics, so that I really have to
worry about welder skill independently of the field,
independently of the repairs and so on? Am I introducing
additional factors that will start pushing the density down
in an artificial way?
SPEAKER: When we did the real elicitations, we
tried to condition every question so that you got an answer
-- for example, we said if you're interested, say, in weld
material, we talked about unrepaired weld material done with
a manual weld and so on and so forth, and they say compare
repair to non-repair.
So things were conditioned and presumably,
hopefully, the experts took account of this conditioning in
their judgments and we never, in the table 5.1 and the
results, we never multiplied -- we only multiplied by one
thing. We didn't multiply two of the expert judgments,
because we didn't have to do that.
DR. APOSTOLAKIS: Did any one of the experts raise
the issue of overlapping? Did they overlap much, some of
them?
MS. JACKSON: Some of them do, but in the backup
slides, on slide 35, that is the beginning of the breakdown
of the quantitative and the qualitative characteristics. So
in the end, we're only using the numbers from the
characteristics that we were able to get exact numbers from
the experts for.
Specifically, that was for the product form, the
weld processes, the flaw mechanisms, the repairs, the flaw
location and the flaw size.
So the majority of the characteristics, we don't
have any -- we're not going to use numbers. In the first
few elicitation sessions, we did ask the experts to compare
welder skill for the different weld processes and finally
some of them said, you know, that is such a human factors
related issue, you can't pinpoint a number, same for
inspector skill.
So some of the things, we're not going to use the
numbers. It will be used when we do the uncertainty
studies, but --
DR. APOSTOLAKIS: I thought this kind of
discussion will be beneficial if you were to insert it into
section four, where you discuss the issue.
MS. JACKSON: Section four, okay.
DR. APOSTOLAKIS: So the qualitative issues were
not used.
MS. JACKSON: Well, they -- I have a slide here.
Let me --
SPEAKER: The qualitative issues were not used to
generate any of the numbers.
MS. JACKSON: If we can go to this, this is a
distinction that we came up with between the two different
types of characteristics.
The quantitative are the ones where the experts
were actually able to provide numerical comparisons and we
will be able to get some records.
We're still receiving some construction records
for some of the vessels that PNNL has. And these
qualitative characteristics, the experts were unable to
meaningfully quantify or the records are unavailable. So in
essence, we're not going to be able to get any numbers for
those qualitative characteristics.
DR. APOSTOLAKIS: What do you mean necessary
records are unavailable?
MS. JACKSON: Like for some of the things on
welder skill, there's really no records for welder skill.
There is no way for you to quantify that on the welder
skill, because that varies so much from welder to welder,
what day; if, five days before a Super Bowl, welder skill
goes down. There's just so many factors, it's hard to
pinpoint exact numbers to compare welder skill for a
submerged arc versus an electroslag, the automatic
processes.
So that's what we meant when we said the necessary
records are unavailable.
DR. KRESS: Measure of welder skill is how many
flaws there are. It's kind of strange trying to use the
same measure to determine the outcome.
MS. JACKSON: Let me put these two slides up. I
think in your handouts, they are in a different format, but
this shows them a little larger.
This is the sheets that we used when we were going
for the elicitation sessions with the experts. So I'm going
to do this as an example.
We asked them about the product form and the
product form was broken down into four different parts;
forgings, plate, the cladding and the weld metal.
So we asked the experts, we said which one of
these is most likely to have a flaw, using that definition
of a flaw that I showed you earlier. So we asked them to
write them and for this one, for example, say this was one,
weld metal had more -- more likely to have a flaw, one, two,
three and four. I'll just use that arbitrarily.
And then after that, we asked them to compare,
okay, so weld metal has the most number of flaws. Compare
the weld metal to the cladding. Which would have more
flaws, the weld metal to the plate, and the weld metal to
the forgings.
Then after that, we asked them -- we added this
late, because initially flaw size was not in here, but we
wanted to know would you have a variation of flaw size and
what effect the fabricator would have.
We had three major fabricators and Combustion
Engineering, Babcock & Wilcox, and Chicago Bridgeni.
Chicago Bridgeni, most of their vessels were partially field
fabricated.
So a lot of information that we had received
before eliciting the experts for the field fabricated
vessels were not fabricated as well as shop fabricated
vessels, and we found that not to be true, and we actually
finished the elicitation process because even though the
vessels were finished in the field, a lot of them were
partially shop fabricated and we actually had two experts
who actually worked with Chicago Bridgeni and one person was
the actual welding inspector and we found out that they
compensated for a lot of the problems that you would have in
the field with the environmental conditions and things like
that.
So that's how we got the numbers. We went through
this for each one. We went through the weld processes. We
had five different processes.
This is one of the areas that was since revised
and went through the elicitation process because a few
experts told us that you need to break this down, because
there were manual and automatic types of cladding, and we
needed to break that down. So that was actually broken down
further.
You had many, many different types of flaw
mechanisms for base metal and for weld metal. So we went
through this and this is where we began to find problems
with the experts. They said the weld procedures were -- a
lot of them -- most of them were qualified, so the weld
procedure should not have that much effect.
So that's where we decided we had to break down
the characteristics into the quantitative and qualitative,
because we couldn't actually get numbers from the experts.
The next two slides just state some of the
conclusions from the expert judgment process. They feel
that it can be done, but it's going to have a wide range of
uncertainty. The flaw density of base metal is
substantially less than for weld metal. The number that's
been used for many years is that the base metal had ten
percent of the flaws of weld metal and the basis for that
was a phone call between Mike Mayfield, Spence Bush. Now we
have some additional data, so we have a basis for that.
Discontinuities in the cladding, that was another
issue that we discussed with the experts.
DR. KRESS: When you say weld metal, are you
counting the region around the weld part or just the weld?
MS. JACKSON: The heat affected zone?
DR. KRESS: The heat affected zone.
MS. JACKSON: No, the heat affected zone, that was
a big problem because it still is actually base metal, but
it's been affected by the heat from the weld. So we include
it as base metal, but take into account that it has been
altered.
DR. KRESS: Okay.
DR. APOSTOLAKIS: I guess you're not getting into
the actual processing of the numbers.
MS. JACKSON: Lee is going to go into that a
little bit.
DR. APOSTOLAKIS: So we should hold off. Are you
going to use the methodology, slides on methodology?
MR. ABRAMSON: Yes.
MS. JACKSON: This is another slide, the last
slide, with some of the conclusions from the experts.
The issue with the large flaws, most of those
should be detected NDE. This discusses two of the
qualitative characteristics, the welder skill and the
inspector skill, and the weld processes are an important
factor in the introduction of flaws.
SPEAKER: When you say NDE, do you mean --
MS. JACKSON: The UT.
SPEAKER: The UT rather than the radiography. But
not all the vessels were UT, right?
MS. JACKSON: They were all -- final to being put
into service, they were all given a 100 percent UT, we
understand, from the experts, prior to being put into
service, either before the actual shell courses were welded,
but we do understand that there was 100 percent UT of the
vessels prior to being put in service.
It may not have been that --
SPEAKER: Even though it only went into the code
at a somewhat later time then.
MS. JACKSON: Right. And it wasn't the extent of
the UT exams that are done now, because these were done so
long ago, but there were UT exams done on the vessels.
MR. HACKETT: I think, Debbie, if I could. This
is Ed Hackett again. I think maybe the more correct
statement would be to say that they all received 100 percent
volumetric exams and maybe the volumetric was a combination
of radiographic testing and ultrasound.
But, of course, given the vintage of when some of
these vessels were fabricated, I think UT was, as Debbie
pointed out, nowhere near in the kind of state it's in today
in terms of the level of advancement.
Plates were typically UT'd. I know if it's a
plate fabricated vessel, as part of the certification for
basically nuclear QA coming out of Lukens, they would have
probably UT examined the plates.
The final composite structure of the reactor vessel,
probably, you could say for sure it received 100 percent
volumetric exam and that's probably, at that point,
restricted to the welds and the adjacent areas. And that
was more than likely, with the early ones, majority RT and
then maybe supplemented by UT, because we are aware of some
of the vessels and it was an issue with the BWRs that some
of them did not receive those level of exams that we would
have liked to have seen and that was -- the committee maybe
remembers the issue, Debbie was involved in this and so was
Lee, over the inspection effectiveness of the
circumferential welds in the BWRs.
And part of the issue there was that some of them
had never actually received one that people would have
agreed upon was a reasonable inspection. And then you got
into the question probabilistically of how important is that
anyway and the industry demonstrated fairly convincingly,
for circ welds in BWRs, that it really didn't matter a whole
lot, is what it boiled down to, because these things were
pretty well made from the beginning, a lot of the things
that Debbie and Lee have been discussing.
So I think I would probably say that is --
MS. JACKSON: That was one of the issues the
experts brought up. The NDE that was done during the period
of the Marshall distribution, it basically picked up larger
flaws. So just the quality of the NDE is a question when
you talk about the final vessel inspections.
SPEAKER: Prodigal gives you a fair amount of
credit for the x-ray, the radiography.
MS. JACKSON: Yes, it does. It does. I just have
some concluding remarks regarding the whole process.
We still have a lot of work left to do. The
report that you have that was dated in July, that's under
revision, and one is coming out at the end of the month and
then it will be revised again periodically before the final
one comes out at the end of next year.
DR. APOSTOLAKIS: But you don't have access to the
experts anymore.
MS. JACKSON: No, I do. I still --
DR. APOSTOLAKIS: The ones that are alive.
MS. JACKSON: -- discuss with them. That's an
issue.
DR. APOSTOLAKIS: So you do. So you maybe can get
some more information.
MS. JACKSON: Some of them -- yes, because I've
had some -- we weren't able to get an expert who was from
Lukens, but I do have a gentleman who retired from Lukens
who does answer questions that I have occasionally.
But some of the people just didn't want to
participate or were retired, and they were retired and they
didn't want to have to go through the process.
Lee?
DR. APOSTOLAKIS: So you did conclude that the
expert judgment process is complex.
MS. JACKSON: Successful and complex, yes.
MR. ABRAMSON: I'm going to talk about the flaw
distribution methodology, and that's in contrast to the --
this is what was intended. It was an upgrade to the
Marshall distribution.
The Marshall distribution essentially combined all
the various factors and came out with a distribution. What
we've done is to separate out these things and I'll talk
about, of course, how it was done and there are certain
advantages to this.
There are essentially three elements to the
distribution. One is the flaw densities and two is the
volumes or areas, and each of these is plant-specific. Then
we have the distribution of crack depth, given that there is
a flaw. So it's combined into these three elements and this
is -- we're treating this so far as generic.
DR. KRESS: Are all flaws treated as cracks when
they get around to doing the fracture mechanics?
SPEAKER: I can give that one a go. I don't think
that's fair to Lee.
MR. ABRAMSON: Yes.
SPEAKER: The report that Dr. Apostolakis was
referring to, at least on PVRUF, there were distinctions
made between volumetric and planar. So from the detailed
NDE, where the defect was considered to have volumetric
characteristics, those were screened out.
So in other words, if you had, in the idealized
sense, a spherical defect of some sort, that was not
considered to participate. The others were just assumed to
be crack-like.
MR. ABRAMSON: This is an outline of the
methodology. The ultimate goal of the distribution, at
least as far as the computation is concerned, this is what
will be input to the FAVOR code, is to get two numbers, the
number of small flaws and the number of large flaws. We do
everything for small and large, because the experts have
told us and, actually, we know from our own experience and
knowledge, but mostly the experts have told us that there is
a difference between small and large flaws. There could be
a difference.
And it's defined in terms of the bead thickness.
A small flaw is one such the crack depth is less than the
bead thickness, the large is larger than the bead thickness.
Now, everything is -- distribution is dependent on
three characteristics of a weld. The first is the product
form, the second is the weld process, and the third is the
repair state.
The weld process we considered was estimated
manual welds, automatic welds, electroslag, when it's
appropriate, and then for the cladding, single and
multi-wire, and repair state is repaired and unrepaired.
So the distribution we're going to get is going to
be dependent upon the various combinations of these.
Then what we have is we have a density of small
and large flaws as a function of the product repair state
and it's per unit volume or area. Areas we use for
cladding. For unit area, everything else is -- the weld
metal is per unit volume. And you do the obvious thing.
We first have, we have N-sub-S, which is just the
number of small flaws. This, of course, is going to be a
sum of products. We have particularly density as a function
of the various characteristics multiplied by the appropriate
volume or area for that point.
So that takes care of the first two parts,
aspects, and the last one, of course, is the density of
flaws and they're defined as G-sub-S, these are the CCDFs,
the complimentary distribution functions. For small flows,
the probability of the crack depth is larger than whatever
the quantity of X is, define those.
And then putting all this together, each GFC, this
generalized flaw distribution, is the product of the number
-- actually, that should be the sum of -- oh, each one is
the product of the number of flaws in the corresponding
crack depth distribution.
So we have -- this is what I started out with in
the first slide. This is a number larger, flaw larger than
X, it's the number of small flaws multiplied by the
probability of it being larger than X, given there's a flaw,
and just pull all this together.
Now, what we have in this -- and this has been
revised based on additional input and commentary from PNNL.
So this is not example -- this is not what you got here, but
this is the latest that we have now.
This is the PVRUF distribution, because it's based
on the PVRUF examination which PNNL did, and specifically
for the volumes and areas. So this is, as I said, the
distribution, of course, is going to be plant-specific and
the plant -- the vessel we're using is a PVRUF vessel here.
Let me just go over this. First of all, here we
have the combination of product form, weld process and
repair state. So we have this for relevant ones here,
first, for the weld metal and plate and, secondly, for the
cladding. We divided this up here.
Then here are the measured PVRUF volumes in terms
of cubic meters for these quantities. There were no -- this
is the plate manual repair. There were no repairs that
we're aware of in the plate. That's why this is zero rather
than a dash. And similarly, for the cladding. Again, here,
there was no multi-wire in the cladding. So that's why the
zero here.
And this is unknown, they're still working on
this, I believe. There's a possibility that they haven't
finished that yet.
So that this column here is the plant-specific, in
this case, PVRUF-specific numbers.
Now, the densities, that is based on the PVRUF
data and also on Shoreham data. Now, this may very well
probably -- I'm sure it will change, because the PVRUF data
has been validated. The Shoreham data has not been
validated. So as we go through with this, it will be
revised, but this is our best estimate on it so far.
I should also say, too, talking about best
estimates, the numbers here are best estimate values. They
are based on data, where it's available, because data trumps
expert judgment all the time, as far as I'm concerned, and
we only use the expert judgment when it's necessary and we
don't have the available data.
So we're just using the actual data, the point
estimates, if you will. And then the expert judgment, and I
can discuss this later, if you like, we're using essentially
the median values. We're using a best estimate for the --
DR. APOSTOLAKIS: But eventually there's going to
be number two.
MR. ABRAMSON: That's right. No. Absolutely.
We're very definitely going to use the uncertainties and for
the data, we'll be able to have it statistically based. The
expert judgment, I'm not quite sure how we're going to do
it, because the experts differed a lot among themselves.
So we have variability, even when we use their
best estimates, but we also elicited low values and high
values for everything that we elicited. So we do have a lot
of information that we can use to construct an uncertainty
distribution, and we certainly are going to do that.
DR. APOSTOLAKIS: Was the Marshall distribution
based on expert judgment? I don't know.
MR. ABRAMSON: Yes. My understanding of it is
yes. Very much -- I think it was expert judgment and, of
course, the available data at the time. That's right,
definitely.
DR. APOSTOLAKIS: You make the observation here
that the density of flaws in the PVRUF and Shoreham vessels
is significantly greater than predicted by a Marshall
distribution.
So I guess that's an indication that the experts
were optimistic. Is that observation going to affect
anything you're doing?
MR. ABRAMSON: Depends on ultimate -- I mean,
affect it, all of this is going to be input into the FAVOR
code, which will ultimately calculate a probability of
vessel failure.
DR. APOSTOLAKIS: I understand that. But
regarding the density, is it possible that your own experts
will be optimistic just as those who helped the Lord?
MR. ABRAMSON: Well, it's certainly possible. We
did not ask any of them for density numbers. All of these
are based on data.
DR. APOSTOLAKIS: You're modifying them.
MR. ABRAMSON: We're modifying it, that's right.
DR. APOSTOLAKIS: So those factors --
MR. ABRAMSON: It's possible. Well, it certainly
is possible, but --
DR. APOSTOLAKIS: I mean, there is no way you can
take this into the ratio, I suppose. I don't know.
MR. ABRAMSON: You have to look at the whole
process. When we elicited the experts, we not just elicited
the opinion. Matter of fact, in a sense, that was the least
time spent on that. We wanted to know their rationale for
all of this and in the report itself there's going to be a
much more fuller summary of the rationales for all of this.
There was also a significant amount of
disagreement among the experts and so on.
So the only thing I can say is -- and there was
certainly significant uncertainty and insofar as the
uncertainty is going to affect the answer, that will
certainly be reflected in it. In some cases, it won't
matter.
DR. APOSTOLAKIS: Speaking of uncertainty, I have
a couple of comments on the report. You have used, in this
calculation, as we just said, the mid value of the range of
the medians.
MR. ABRAMSON: Essentially. Or the median of the
mid values. However you want to look at it.
DR. APOSTOLAKIS: Median of the mid values. Now,
I hope you're not going to define medians and high values
and low values when you actually do your uncertainty
analysis. I think the accepted way of doing it now is to
actually have the distribution of each expert, put them all
on the same plot, like NUREG-1150 did or the Shack report
did -- not this Shack -- it's S. Shack -- and you select a
point on the abscissa and you go up and you find all the
experts take the mean, and that will give you a distribution
of the fraction because of the expert assessments, or you
can analyze it in a different way and there is a long
discussion in the appendix of that seismic report.
If you want to single out the variability -- the
expert-to-expert variability, I don't know what you're going
to do with it when you go to FAVOR, but maybe that would be
an additional insight.
But what I think -- and the whole idea behind all
this -- this is the idea of equal weights. You are giving
equal weights to the expert distributions because, as you
make a point here on page 14, the ensuing discussion served
to ensure a common understanding of the issues and the data.
Since you had this feedback, then there is no
reason really for you to give different weights from
different experts, which is really --
MR. ABRAMSON: We have no intention of doing that.
Absolutely not.
DR. APOSTOLAKIS: But I think you should give
equal weight to their distribution, not to the -- don't take
the medians and add them up and divide by 17.
See the difference?
MR. ABRAMSON: I'm not sure that a distribution
has any meaning here, because all we're asking is low, mid
and high values. I don't see that -- it doesn't make any
sense, to me, to --
DR. APOSTOLAKIS: You would have to make some
assumption regarding the distribution. I mean, is it a lot
-- normally, these things are --
MR. ABRAMSON: I don't think so. I don't think it
has any meaning. All we did is we asked -- when we asked
the experts for the low, mid and high values and we went
through a training session, I think they all understood what
we were asking.
The mid values, of course, are the approximately
median and a low value is one such that there's only a five
percent chance, in their judgment, that you could be lower
than that, and a high value is only a five percent chance
you could be above that.
DR. APOSTOLAKIS: Yes, but the fact that you don't
have that piece of information probably doesn't justify
adding the medians and dividing by 17.
MR. ABRAMSON: I'm not adding the medians. I
don't --
DR. APOSTOLAKIS: All I'm saying is in the future,
if you do that, it will not be consistent with what the
community thinks.
Now, you don't have the information of the
distribution between low, medium and high, but maybe you can
put something there and speculate and then see how the
summation comes up.
I mean, you will have to do something anyway,
because you don't have sufficient information.
MR. ABRAMSON: I know.
DR. APOSTOLAKIS: All I'm saying is there are two
major studies, 1150 and the other one, the senior seismic
hazard analysis committee report, which really spent a lot
of time on these issues. They both recommend that when you
are reasonably satisfied that the experts deserve equal
weight, then you do what NUREG-1150 did.
You have the variable, you put the distributions,
and then you go up each point and you add up the
probabilities of what the experts gave you and find the
value, and that gives you the composite uncertainty. And
there are other ways you can analyze it, too, but this is
the accepted way.
This is just a suggestion for the future that you
may want to consider, because you're on the right track. I
mean, you had this discussion of the issues and assuring a
common understanding. Then you can say because of that,
this is what we're going to do.
Let's see. Now, for these purposes here, taking
the mid value is just a representative example.
MR. ABRAMSON: We're just trying to get a ballpark
estimate at this point, that's right.
DR. APOSTOLAKIS: Okay. I guess that's it for the
time being.
MR. HACKETT: This is Ed Hackett. I'd like to add
a comment on what Professor Apostolakis mentioned on the
density, so as not to cause undue alarm. Several factors
come into play there. The issue with saying this
distribution has produced a much higher density of flaws
than Marshall, first off, shouldn't be surprising, because
what you're seeing is advancement in the state-of-the-art of
the NDE.
Then you could direct your attention to the boxes
over to the bottom right on Lee's chart there and that's
kind of illustrative right there. You look at the number of
small flaws and you see this 22,000 number and then you get
down to large flaws, which are getting closer to the
category of what would participate, as Terry would put it,
if you were looking at FAVOR probabilistically in a PTS type
transient. It's going to be a much, much reduced number.
So the fact that you're seeing, which is mainly
focused at the clad-base metal interface or in the weld
metal, is not an alarming thing. It is one of the things
I'd just like to leave everybody with.
DR. APOSTOLAKIS: I think it would be helpful and
useful to have these comments in the report, because this
statement is hanging there.
MR. HACKETT: That's one of the reasons I brought
it up, because it has tended to alarm some people and it's
really not the case. Most of those flaws are not going to
-- the vast majority of that number there that's got 22,000
is not going to participate significantly in response to a
PTS transient.
MS. JACKSON: I think in one of the documents that
we're going to send you, you'll see that a lot of the flaws
are just very, very small and they have no interest, no
interest at all.
MR. ABRAMSON: The details are going to be given
in the report and these densities are based where they were
applicable, and, in many cases, they were based on data from
the PVRUF, both from the Shoreham and the PVRUF flaws.
And when they weren't, we augmented it with expert
judgment.
SPEAKER: Well, I mean, let's be specific. The
welds are based on data, the others are based on expert
judgment, right?
MR. ABRAMSON: Let's take a look. Well, there's
also a question of repair and non-repair. I think, yes, I
would say the welds were the expert judgment, where the --
SPEAKER: The repair is probably an expert
judgment.
MR. ABRAMSON: Where the expert judgment was used
was -- it was in the plate, that's right.
DR. APOSTOLAKIS: And this is the expert judgment
--
MR. ABRAMSON: Now, the plate --
DR. APOSTOLAKIS: The 17 experts?
MR. ABRAMSON: Yes, that's what I mean. The
expert judgment, modified. That's right. They do not have
very much plate data, but they are getting some. So this
will be replaced by the plate data once we get it, and the
cladding, also, I think, was used to some extent in the
expert judgment.
And then, of course, to fill this out, we just
took these estimated densities by the measured volumes and
multiplied and that's where these came out.
Now, there were a very large number of small
flaws, but we fully expect that they really are going to
contribute essentially nothing or very close to nothing when
it comes down to the fracture mechanics in the FAVOR.
The ones that, of course, will contribute will be
the large flaws.
Now, we divided this table into two parts. The
bottom half is the cladding and there large flaws can be
most of the thickness of the cladding, which is six to eight
millimeters. So, again, we feel that that will probably not
contribute at all once it goes through the fracture
mechanics.
So the ones that will contribute will be the large
flaws here and, again, we emphasize, this is just a
preliminary estimate base that we have now. A vast majority
of these were from the weld metal manual, repaired, and
repairs are manual.
And this is what we've learned from the experts,
that repairs are much more likely than non-repair for metal
to have flaws in them. So that's what is driving this.
And we do have data on this, as well. I think
this was based on data because there were some repaired
regions here, like we see here.
DR. APOSTOLAKIS: So the density of large flaws is
96.
MR. ABRAMSON: No. The number, this is the number
of flaws. This is the estimated number of large flaws in
the entire -- in the PVRUF vessel, the part of it's subject
to PTS. That's the estimated number. A total of 96 large
flaws in the valve line.
DR. APOSTOLAKIS: So what will be the input?
MR. ABRAMSON: Into FAVOR?
DR. APOSTOLAKIS: Yes. Ninety-six?
MR. ABRAMSON: The number will be 96. Of course,
it will be distributed to location, and Terry will go into
this in detail.
But if we used this, if we ran FAVOR tomorrow, we
would say, yes, you start with a total of 96 flaws in the
valve line region.
DR. APOSTOLAKIS: So you start with 96 flaws.
MR. ABRAMSON: Right, exactly. Of all sizes, I
should say. This is the total number of large flaws. And
you would apply the distribution, which I'm coming to, to
get the specific sizes of those.
DR. APOSTOLAKIS: Now, if I go back to the
Maryland paper on uncertainty, that figure six, it says
flaws exhausted. What does that mean? They will do it for
each of the 96?
MR. ABRAMSON: Yes.
DR. APOSTOLAKIS: Each of the 96. Why? I mean,
they have a distribution, don't they? I don't understand
that. Anyway, we'll discuss that when the time comes.
Flaws exhausted, you do it for every single one? You're
going to take the probability that there is a flaw there and
the distribution of the size and just do it? I don't
understand what it means to exhaust the flaws.
You're given the total number, you have a certain
volume, right?
SPEAKER: You're talking about in the context of
the University of Maryland paper, flaws exhausted. What
that means is each vessel, let's say, has 96 flaws, if
that's what the case is. You calculate the probability of
fracture for each one of those flaws and then the
probability of fracture for the entire vessel is kind of a
summation process.
DR. APOSTOLAKIS: How do they differ?
SPEAKER: Well, flaw number one, you're going to
first sample it to find the size of it, and it may be in a
different location.
DR. APOSTOLAKIS: Each flaw may have a different
size.
SPEAKER: Yes. As well as be located at a
different part of the belt line region.
DR. APOSTOLAKIS: All right.
SPEAKER: As well as be located at a different
location through the wall.
DR. APOSTOLAKIS: Anyway, we'll discuss that in
November.
DR. KRESS: And if you get enough samples, we'll
just sample 96, you sample thousands to cover that.
SPEAKER: However many flaws are in the vessel,
that's how many you sample.
DR. APOSTOLAKIS: See, if you postulate that you
have 96, then you have to do it, right? But I don't know.
That's new to me.
DR. KRESS: I would have thought --
DR. APOSTOLAKIS: You're postulating that there
are 96, no matter what, and now you worry about where they
are and what the distribution of the size is.
DR. KRESS: Yes, but if you just take one flaw and
then fix its location and size by sampling, it seems to me
like 96 samples is not enough. You have to -- you don't
cover the map that way.
DR. APOSTOLAKIS: I guess that's why it's
important to understand what sampling means. Is it from the
aliatory -- is it epistemic, aliatory, how do they come
together, but I guess we'll have another subcommittee
meeting on this.
All right. Back to you.
SPEAKER: Maybe a point that's not clear for
Professor Apostolakis' question. Of course, you're going to
be doing many vessels, perhaps a million vessels, each with
the 96 flaws.
DR. KRESS: That's what I'm --
SPEAKER: Each one of the vessels has a certain
number of flaws and you're doing many, many vessels.
DR. KRESS: That covers many vessels, yes.
DR. APOSTOLAKIS: You select the vessel.
DR. KRESS: It's the way you phrase it.
MR. ABRAMSON: As I said before, this is also a
modification. To show that I meant what I said, I'm going
to modify it right now. Actually, this was a slide taken
from the presentation we made in August, but subsequent to
that, we've modified it and as I said, the current thing is
going to appear in the report, which is going to be out, I
guess, in a couple of weeks or so.
Where it's going to be modified is that this --
the large density, this is --
DR. APOSTOLAKIS: I don't think it matters.
MR. ABRAMSON: -- 700, densities. The numbers --
this becomes 40, the number was 40 here, so the total
becomes 66. So it's somewhat less than this. Don't rely on
this as far as -- and it may be modified -- it's a new table
and, also, since the FAVOR runs are not going to start for a
number of months, the numbers that we put into it, as we get
more information from PNNL, we certainly are going to modify
the inputs for FAVOR. So that may change it further.
But right now, it's somewhat less than 66, rather
than 96.
SPEAKER: Which is certainly different than 2,581.
MS. JACKSON: Right.
MR. ABRAMSON: That's right. It keeps going down
apparently. Now, the final part of this is the -- this is
CCDF for the large and small flaws and here is what we're
using right now, what's available right now.
This is based on the large and small flaws that
were observed by PNNL.
END TAPE 2, SIDE 1.
TAPE 2, SIDE 2 FOLLOWS:. BEGIN TAPE 2, SIDE 2:
Most of these came from Shoreham, which has not
been validated, and some of them, maybe about a third or so,
came from PVRUF, which has been validated.
But I put them all together and we get this
distribution, empirical distribution, which is based on
something like 64 total flaws all together, which, to my way
of looking at it, is remarkably smooth.
This has not been smoothed, by the way. We just
connected up the points.
DR. APOSTOLAKIS: So this is large flaws anywhere.
MR. ABRAMSON: That's right. Large flaws -- well,
in the weld metal, that's right. Repaired, non-repaired
material, we just threw them all together. That's right.
And the assumption we're making, working assumption we're
making right now is that this is a legitimate thing to do.
We can combine flaws from all different kinds of
weld metal and so on made under different welding conditions
and, in other words, a large flaw is a large flaw, as far as
this is concerned.
It doesn't matter what material it was in as far
as the crack size distribution is concerned. So this gives
us the power of doing that. That's how we're planning to
use it at this present time.
And there's a --
DR. APOSTOLAKIS: So ten percent chance of having
a flaw greater than ten millimeters. Wow.
MR. ABRAMSON: That's what the data showed. I
mean, this is based on the data, that's right. This is
based on the data. Ten percent of the large flaws were --
that's right, exactly. Which isn't a very large --
remember, George, this is -- we're talking about maybe six
flaws all together.
DR. APOSTOLAKIS: How many?
MR. ABRAMSON: We're talking about maybe six flaws
all together, but that's how it's coming out.
DR. APOSTOLAKIS: So if the process was faulty and
there is a large flaw, then probability that it's really
large is not negligible. What saves you is that you don't
have too many of those.
MR. ABRAMSON: And, of course, there is a
significant amount of uncertainty in this.
DR. APOSTOLAKIS: The process.
MR. ABRAMSON: A significant amount of uncertainty
in this distribution and when we do the final analysis, that
will be reflected in that.
DR. KRESS: Is that for the base metal?
MR. ABRAMSON: It's for flaws found everywhere.
Actually, I don't think we have any flaws in the base metal
because they didn't inspect any of that yet.
DR. KRESS: Okay.
MR. ABRAMSON: This is just flaws in the weld
material.
MS. JACKSON: A small area.
MR. ABRAMSON: Only a small area.
DR. KRESS: That's why the distribution goes below
five millimeters.
MR. ABRAMSON: That's right, yes.
DR. KRESS: Because it's bead size rather than --
MR. ABRAMSON: Bead size, right. We did that
definition. Exactly, that's right.
MR. HACKETT: I guess the other comment I would
add -- this is Ed Hackett -- is this is not -- I think Lee
stated this earlier. This is also not addressing location.
So it could be that even out of the six, in all the greatest
likelihood, they're not located on the surface, in which
case you may not have any participation at all, depending on
where these flaws are located.
SPEAKER: How does this compare with what you find
when you do UT inspections in the field? How many one-inch
long cracks have you found?
MR. HACKETT: This is Ed Hackett, again. I know
there may be some others in here who could comment on this,
too. My understanding is nothing in that range has been
found, that I'm aware of. Bob Hardy is here.
MR. ABRAMSON: We're predicting six per vessel.
MR. HACKETT: I think what you're looking at is
the statistics of the process and then, also, we have not
gotten into -- and Lee and Debbie haven't included this --
how good are the inspections versus what was done for PVRUF
and Shoreham. Obviously, these are laboratory conditions
and they're able to destructively verify what's there and
what isn't.
Of course, you can't do that in the field. The
NDE is better than it's ever been. But I don't believe --
maybe others in the room can comment. I'm not aware of
hearing anything in that kind of size range that's come from
a field inspection that would be in a surface location.
I think there have been isolated cases where
larger flaws, like on the order of multiple millimeters,
have been located at different points in the depth or maybe
towards the outer surface, but I'm not aware of any.
I don't know if you are, Debbie.
MS. JACKSON: No.
MR. HACKETT: Not in field inspections.
MS. JACKSON: On the large flaws that we've been
finding in the PVRUF and Shoreham are in the repaired area.
There was a large one that we found in Shoreham that's about
30 millimeters, but it hasn't been validated. So we don't
know.
According to proximity rules, if it's a cluster of
many small flaws, but the largest one found so far in PVRUF
was 17 millimeters.
MR. HACKETT: I would also add, though, Bill,
you're right in that if we do enough of these and we're
right about what we're doing here in the lab, eventually we
should find these things. I think it's just a question of
the statistics of the process and how good is the field NDE.
MS. JACKSON: Right.
MR. HARDIES: This is Bob Hardies, from Baltimore
Gas & Electric Company. The largest flaw so far that's been
validated, the 17 millimeters, was a cluster of small
volumetric things.
So really everything so far that's been
destructively examined that's been larger than ten
millimeters are really little porosity clusters.
MS. JACKSON: Right.
MR. ABRAMSON: Three of those, and here are 14,
21, and 32, these are from Shoreham data, which has not yet
been validated. Also, these large flaws bigger than ten,
three of them -- some of them were repaired, but others were
non-repaired.
We had -- again, this is not validated. It's 21
and 32 came from non-repaired material. Again, this is all
subject to possible revision once they validate the data.
And this is the CCDF for small flaws and this, I
believe, is based only on the PVRUF, I believe. This is a
lot choppier, but, again, this is what the data show at the
present time. Again, I repeat that we don't expect the
small flaws are going to contribute in any significant way
to vessel failure.
So this is of interest, but it's not really going
to affect the bottom line as far as PTS is concerned.
Now, how is this going to be used in the FAVOR
code? There's a little bit more detail here.
First of all, we have large flaws and small flaws
and we have weld material and plate material. We don't
expect the cladding to contribute anything significantly,
although certainly we will put it in, but we don't expect it
to contribute anything very significantly.
And what will the -- what the actual input will
be, we'll take the total number of large flaws, in this
case, it's the revised number of 66, and then we'll apply
that distribution to it and come up with the specific
X-sub-I, those are the crack depths.
So we'll take these 66 flaws, 66 large flaws in
the weld material, a certain number in the weld material,
whatever the number is, and a certain number in the plate
material, whatever that total number is, and then we'll just
assign numbers from the large flaw distribution.
These will then be a set of numbers, a set of
crack depths, and this is the weld large flaw and so on.
And similarly for small flaws.
So the input to PVRUF will be the specific; that
is, specific in terms of their crack depth. That will be
the input to PVRUF and then FAVOR, and then FAVOR will take
it from there, locate them and so on.
DR. APOSTOLAKIS: This will be both the aliatory
and epistemic component.
MR. ABRAMSON: I don't know if that is here.
DR. APOSTOLAKIS: You have a distribution.
MR. ABRAMSON: We have a distribution.
DR. APOSTOLAKIS: That would reflect the aliatory
and if you have many distributions, then epistemic.
MR. ABRAMSON: Okay. For any -- that's right.
How we're going to do the uncertainty analysis, that's
right. In effect, we could do -- we'll make draws from
those distributions, correct. That's right.
DR. APOSTOLAKIS: So this would have both.
MR. ABRAMSON: Yes. We'll certainly reflect all
of the uncertainty, absolutely.
SPEAKER: Again, I think that this is something
that we'll need to talk to you guys about at the
subcommittee meeting, because I think depending on how you
view the problem, even though you can talk about aliatory
components leading to observed variability, let's say, in
samples or in vessels, when you lock down on a vessel now,
say, you're hypothesizing a vessel with certain
characteristics and I think it's arguable whether the number
of cracks, for example, is aliatory or you could, in
principal, find them and characterize them, which is a
function of how this is being used in the model.
So that's, again, worth, I think, talking about
when we get together.
DR. APOSTOLAKIS: Okay. See, the problem is that
if you were talking about certain conditions which are from
all different plants, then, of course, it's aliatory because
you pick one plant.
SPEAKER: That's right.
DR. APOSTOLAKIS: Now, if you have one, though,
that distribution becomes subjective.
SPEAKER: That's right. That's how we're viewing
it right now.
DR. APOSTOLAKIS: But the problem is even if you
pick one and you find, say, ten flaws, they will probably
have different lengths, and if you don't have the aliatory
element, then you will probably assume that all of them are
of the same length.
SPEAKER: Again, this process, as I understand it,
is going to say you pick a location -- you effectively,
although it doesn't literally do this, you're going to be
picking a particular flaw with certain characteristics and
those characteristics will include not only the
characteristics of the flaw itself, but the properties in
the neighborhood of the flaw, and that's knowable, in
principal.
DR. APOSTOLAKIS: Okay. We'll discuss that.
SPEAKER: And, George, we would never assign the
same length to every one of those.
DR. APOSTOLAKIS: See, that's the fundamental
distinction between the two, that if you have an aliatory
component, you allow for this randomness. But as Nathan
says, if I understand, will you know enough about this
particular vessel so that you eliminate the random element.
You know the conditions and so on and this and
that, will you have only epistemic.
SPEAKER: The CCDF, this thing, this distribution
is strictly aliatory.
DR. APOSTOLAKIS: That's what I thought. For a
particular vessel, it may not --
SPEAKER: We're going to be sampling using that
distribution. It's how we use that sample in the
calculation that is the point that we're talking about.
There is certainly variability in the flaw sizes
if you look across the population of flaws. How you use
that uncertainty -- this is the plant to plant variability
versus a single plant issue that we look at in the PRA side,
and I see it as the same thing.
Once we start fixing on a particular location of a
particular vessel, and, again, this is all hypothesized,
once you've done that hypothesis, that's what FAVOR is
doing, given that now, is there really going to be that kind
of variability that you're talking about.
And that is where I think -- we have made certain
assumptions in the white paper which try to bring these
things out explicitly. This is why we're saying that this
particular issue is aliatory, this is epistemic, and I think
that would be a good basis to go through the paper.
DR. APOSTOLAKIS: But that would not apply to K.
SPEAKER: Right. That argument does not apply to
K. That's why we said in the paper now we think there is an
aliatory component that needs to be addressed separately
from the things that you're talking about, because there is
the model issue.
DR. APOSTOLAKIS: If you can walk us through a
particular calculation with all these observations, I think
that will be very helpful.
SPEAKER: One thing, again, I need to point out.
I think we've been working through this as part of the
overall PTS analysis and I don't know that we are fixed
right now on the approach that's going to be everlasting
that way.
It's evolving, we're discussing these things. We
will talk to you about where we are, of course, at the time
that we meet. But things certainly can change. I think
we've had a lot of discussions on these specific issues and
how to address them and we do need to walk you through how
we're looking at it now.
MR. GUNTER: Paul Gunter, Nuclear Information
Service. Noting the number of small flaws that you've
noted, I'm wondering if it's too quick of a judgment to
eliminate them as participating in a PTS event.
So could you just give me a quick idea of how you
can make that blanket statement that that many flaws and --
SPEAKER: I'm making that statement based on what
I heard about the likely effects when you put this through
the fracture mechanics code and everything like that, that
small flaws will just contribute very, very little, if at
all, to the probability of vessel failure.
And, actually, I'm not the person -- you need
somebody who can maybe speak more eloquently about that.
MR. DIXON: Terry Dixon, again, from Oak Ridge.
All of the flaws will be input into the FAVOR code. The
small flaws, as well as the large flaws.
Small flaws will be in the analysis and as much as
they contribute, they contribute. But, I mean, we know
certain things just about fracture mechanics. We know that
probably a flaw below four millimeters would never
contribute, but it will be in there.
Essentially, it will be part of the bookkeeping,
but we anticipate that it will contribute very, very little.
So it will be in the fracture mechanics analysis. It won't
be culled out.
MR. ABRAMSON: And just some concluding remarks
about the generalized flow distribution. What it does here
is it combines three areas, three elements, the densities,
which is generic, that would be the flaw distributions.
That's right.
The densities are generic in the sense that they
are not plant-specific. They are certainly product form
specific. They are certainly weld process specific and they
are repair state specific, but they don't depend on the
particular plant that we're talking about. So in that
sense, it's generic.
Crack depth distributions, as I indicated, are
generic, and the plant specific will, of course, have to be
the specific volumes and areas of the weld metal and the
base material in the plant and how much of that was
repaired, how much of that was not repaired, and the weld
process and so on. All of them are very specific about that
plant.
And the generic inputs are based on all available
data and where we don't have the available data, we have to
fill in, then we use the expert judgment from this panel of
17.
So that's the general structure of the generalized
flow distribution, as we have it now.
That's the end of my presentation, if you have any
questions.
SPEAKER: You make the comment in the report that
this thing agrees reasonably well with the Prodigal
predictions. I just wonder what --
MR. ABRAMSON: I didn't make that comment. I'm
not familiar with the Prodigal.
MS. JACKSON: That was some work that PNNL has
done before the PVRUF data was validated. So we've made
some changes to that.
But it was the data from PVRUF and Shoreham was
put into the Prodigal and the predictions came out pretty
close to what came off of Prodigal. That's one of the
things we're going to put in the repot, comparisons of the
PVRUF and the Shoreham.
SPEAKER: I think it's safe to say, Debbie, that
we're also planning to use -- or we don't have data to use
Prodigal runs, as well as expert judgment.
DR. APOSTOLAKIS: See, Prodigal is expert
judgment. That's what confuses me all the time when I see
that.
SPEAKER: It's a different kind of expert
judgment.
SPEAKER: George, it's different expert judgment.
The questions are very different and Prodigal -- I mean, you
need certain inputs into the Prodigal. The Prodigal does
model very explicitly the physical process of welding and
creating flaws and how they would propagate.
MS. JACKSON: And the same with Prodigal. The
Prodigal doesn't deal with base metals. So when we get into
the base metal issue, we can't use Prodigal. It only deals
with weld.
SPEAKER: I just wondered. These numbers seem to
be floating around so much and I assume that depends on
whether you're multiplying by the right weld volumes times
the densities or --
MS. JACKSON: That was another thing, because
initially we had had a different -- I was just concentrating
on the weld volume in the belt line area, that's all. Not
every other thing, the belt line.
SPEAKER: Every other weld.
MS. JACKSON: Right. And then we just recently
got some construction records from PVRUF. So we found some
of the numbers were a little different from what PNNL had.
So hopefully we'll be able to get more information on the
construction records from the fabricators themselves, that's
what we're hoping to do.
SPEAKER: It's comforting to find numbers that
converge. When I see numbers that go from 2,581 to 90 to
66.
MS. JACKSON: Right. Some of those were due to
operator error, also, with the calculators at some point.
SPEAKER: I guess we can start with Shaw's
presentation.
MR. MALLICK: I am Shaw Mallick, the Materials
Handling Branch, and I will be providing a bit on
probabilistic fracture mechanics within the PTS project.
A brief outline is we're going to go provide one
of the major technical areas, the progress made in all those
technical areas, and some concluding remarks.
Here are the six technical areas we are currently
working on. You already have heard about the fabrication
flaw distribution and there will be presentations on
distribution, fracture toughness, improved irradiation
involvement, and the computer code. So this will be a more
explicit presentation.
I will briefly discuss these ones. I'm not sure
if I should go and tell you a little more on that, that you
already have for over an hour.
So I will skip that part of the presentation.
DR. APOSTOLAKIS: I wish you didn't say
statistical representation. Presentation of the
uncertainties.
MR. MALLICK: Okay.
SPEAKER: Well, but this report is statistical.
DR. APOSTOLAKIS: It shouldn't be.
SPEAKER: Well, there is the Oak Ridge report,
which is statistical. Whether it should be or shouldn't be.
I would like to go over the fabrication, number
six, which is the fracture toughness distribution. The
objective here is to provide initiation of fracture
toughness based on expanded ASTME-399, the standard type of
data, and using statistical methods.
And just as background, our latest revision was
developed based on '70s and '80s toughness data and they
were -- not only that, those data were put through an ad hoc
distribution based on lower bound curve.
The Research staff is Mark Kirk, myself and Nathan
Su, in PRA area, and our contractors at Oak Ridge, as well
as University of Maryland, and we are also getting some help
from EPRI and a contractor PEI, Phoenix Engineering
Associate, Professor Marge Natisha, who used to be at
University of Maryland earlier.
Briefly describing the progress made, we will hear
that in about 45 minutes worth of presentation on that in
the later afternoon. Searched and collected additional data
and almost doubled the rate and based on those data, set the
distribution for both different parameters and those
distributions for initiation of fracture toughness K1C, as
well as the K1A.
And one thing in that report that's missing was
uncertainty in the normalizing parameter RTNDT. That is
being looked at separately.
And University of Maryland is assisting in
separating those into epistemic, as well as aliatory
uncertainty, as well as effect of material variability and
model uncertainties. And we expect to have completion by
November.
DR. APOSTOLAKIS: So the variable distribution is
aliatory.
MR. MALLICK: Yes.
DR. APOSTOLAKIS: And then you have three
parameters, A, B, C, each one being a complex function of
delta RTNDT.
MR. MALLICK: RTNDT.
DR. APOSTOLAKIS: RTNDT now will have epistemic
and aliatory itself?
MR. MALLICK: Yes.
SPEAKER: If someone will explain to me again. I
still have problems with whether I'm following a curve or
I'm walking up and down this whole distribution, and I
assume we'll talk about that.
MR. MALLICK: There will be discussion on that
this afternoon.
DR. APOSTOLAKIS: Are you going to have any expert
elicitation exercises in addition to what Lee did?
MR. MALLICK: In the flaw distribution area --
DR. APOSTOLAKIS: You are saying here that
Maryland and EPRI are assisting in model uncertainty. How
are you going to assess the model uncertainty?
MR. MALLICK: They are going through the root
cause diagram, going through what are the basic parameters
building up to the model uncertainty and deciding what are
the uncertainties in those areas.
DR. APOSTOLAKIS: Yes, but there is a model
someplace that is not in a good shape. Somehow you have to
evaluate the uncertainties associated with the predictions
of this model.
MR. MALLICK: Yes.
DR. APOSTOLAKIS: How are you going to do that?
SPEAKER: Let me try that a different way. Short
answer is between us on the staff and University of Maryland
and Oak Ridge, but the question that you're getting to is
the how good of a model is RTNDT at predicting what the --
if we're assuming truth of the situations, we want to get to
the fracture toughness and we get there by using RTNDT as an
index, how good is that as a model.
We have never addressed that explicitly before and
as you mentioned, there are both aliatory and epistemic
components to that. That is going to be addressed -- I
guess I can back up and say this is another area that could
have easily lended itself to expert elicitation. I think
what we're looking at is running up against resource
limitations on being able to do that.
So what we're doing is trying to do that as a
group between the staff and, in this case, University of
Maryland and Oak Ridge.
DR. APOSTOLAKIS: Internal experts.
SPEAKER: Right.
DR. APOSTOLAKIS: But you will do it.
SPEAKER: Yes, absolutely.
DR. APOSTOLAKIS: That seismic -- is Lee here?
SPEAKER: Yes.
DR. APOSTOLAKIS: That seismic report gets you way
out of this, because it defines two ways, two major ways for
doing an analysis using the technical integrator or the
technical facilitator integrator, and depending on the
significance of the issue, you may go with the technical
integrator, which is a less formal way of eliciting
judgments and I think that's what they have just described.
But as for the other stuff that you just
presented, you really did the tier five, because it was
bigger, broader and so on.
So I think there is a lot if information there
that will help you. There are two volumes and -- I don't
know. Do you know which volume I'm referring to?
SPEAKER: Yes.
DR. APOSTOLAKIS: Okay. Because the technical
integrator approach was used without this for Diablo Canyon,
we were told at the time, and it worked very well.
Everybody liked it very much.
You didn't have to go out of your way to bring
experts and fly them over to Albuquerque, usually, and do
these things.
SPEAKER: We did a lot with video conferencing.
DR. APOSTOLAKIS: So there is some merit to that.
You know, if you give something a name, it's automatically
more respectable.
MR. MALLICK: The next area we are looking at is
embrittlement correlation development and the objective here
is to revise the predicted shift in RTNDT using up-to-date
data, as well as the statistical data, and not only that, we
are also trying to -- in the process to revise the Reg Guide
1.99, which two of the three -- there will be a three
document draft developed and we want to have consistency
with that guide, as well.
Then they become part of the rule, as well as a
guide, going in parallel and they are addressing the same
issues.
Currently, we're looking at the correlation for
Reg Guide 1.99, and it's based on earlier data. Again, we
have at least three times more data now on embrittlement
correlation than we had at that time of the data set.
And this is Mark Kirk and Carolyn Fairbanks, and
the contractor for the NRC side is Modeling and Computing
Associates, which is Ernie Leeson. Oak Ridge National Lab
is Randy Nanstadt and his group, as well as University of
Maryland is, again, helping us.
SPEAKER: What is PEAI?
MR. MALLICK: It's the Phoenix Engineering
Associates Incorporated.
Progress made in this case. We have a mean
correlation. End of August -- end of July, sorry, we have a
mean correlation.
DR. APOSTOLAKIS: What is a mean correlation?
MR. MALLICK: Mean is best estimate correlation
and we are trying to -- the next step is to characterize.
DR. APOSTOLAKIS: So much educating to do. So the
uncertainties characterize using the approach that Debbie
described.
MR. MALLICK: Yes.
SPEAKER: That's correct.
DR. APOSTOLAKIS: So among us.
MR. MALLICK: After lunch, we'll have some more
discussion on that.
SPEAKER: This is one that I think it's fair to
say an expert elicitation might have been a benefit here.
DR. APOSTOLAKIS: Again, you don't have to limit
yourself to the --
SPEAKER: The formal process.
DR. APOSTOLAKIS: -- inside people. I mean, you
don't have to have a very formal process and still consult
with outside experts. Sometimes even a phone call. It's
better to have information than not to have it.
SPEAKER: This has actually been the case with
this one, because this actually, in terms of -- the mean
correlation that was developed by the work of Ernie Leeson
and Bob Odette principally was sort of vetted out even in
the 1998 timeframe.
SPEAKER: But when you get an ASTM, you're
essentially getting a certain advantage of opinions.
SPEAKER: And then ASTME-10 committee has had a
lot of discussion, influence, et cetera, on some of the
direction where that's going. So that's been vetted at
least among industry groups and consensus codes and
standards folks, too.
DR. APOSTOLAKIS: Please don't call it mean.
MR. MALLICK: Best estimate.
DR. APOSTOLAKIS: Just nothing. There is no such
thing as best estimate either. That's okay. Best estimate
is better than mean.
SPEAKER: Best guess.
MR. MALLICK: Both in Maryland, the correlation
analyzes the fracture toughness and the specific material,
so we can see the solution to the input, and these
activities are using industry data to come up with the
distribution in terms of distribution for copper, nickel and
phosphorous.
Also, it is to get the local variability. For the
weld case, we have four PTS plants. In this, we have
something like 15 weld heats, with two nickel addition, as
well as 16 plate heats, and the work is virtually internal.
Doug Kornoski, Tammy Samples, and Lea Berser, as well as the
industry to get their data, a lot of data from the industry.
For the weld case, we have some heat distribution
already. They are essentially normal. And we also have
local variability. Welds are presented using distribution
of copper and nickel, as well as normal distribution for
phosphorous.
In the case of plates, the data set is somewhat
limited. So data is limited for the heats in the PTS
plants. So chemistry was taken as heat estimate and we
didn't have as good a distribution as we had for the weld
material, but the plates are much more uniformly fabricated
and things like that. So they were much less as it would be
in this case, the effect of variability, that is. So,
again, plates, we have limited data we obtained and we need
to develop a solution on that as well.
SPEAKER: By looking at this variation, are you
going to change the margin type terms that you would usually
use in a Reg Guide 1.99?
MR. MALLICK: They will go as a -- the
distribution will go in the analysis. We probably do not
have a margin.
SPEAKER: You'll replace the margin with this
distribution.
MR. MALLICK: Yes. The next major area we are
working on is the neutron fluence calculations and our
objective for this activity is to determine an up-to-date
end of life fuels map for the plants, all the four plants we
are looking at, using currently available cycle by cycle
data of the fuel loading, as well as the plant data and also
to have some kind of estimate for uncertainty in the fluence
calculation, as well.
And we are using draft dosimetry guide 10.53, I
think this will be coming soon, as well as corresponding
NUREG report. That staff is Billy Jones and we're getting
help from Brookhaven National Lab on that.
Plants on-line so far, all the three plants,
Oconee-1, Palisades and Calvert Cliffs have been analyzed.
We also had analyzed Robinson, but it's not in the running.
So we are replacing it with Beaver Valley and we are just
receiving the plant data from Beaver Valley, we have to look
at to what extent we have to perform analysis on that.
And Brookhaven has performed very defined grids
for actual circumferential, as well as the radial direction.
For example, here is the example given for Oconee, Palisades
and Calvert, actually is 218, and the corresponding
circumferential is 60 nodes. Similarly, we have a very
refined grid going in.
Now, Brookhaven also has calculated some kind of
uncertainty in the fluence calculation and for each of these
three plants, one sigma in fluence is about three percent of
the mean value.
And we are internally looking at do we need to
perform some kind of modeling interaction among these
various fluence parameters, such as vessel damage or nuclear
cross-section, they are the major contributor for the
uncertainty. So we're going to go look at the interaction
between them and that may have some effect on this number of
13 percent answer.
SPEAKER: Your comment on the non-linear
interaction of parameters, you mention the core inlet
temperature. Is that a strong parameter?
MR. MALLICK: Those parameters are -- it's five
percent of the mean or something like that is contributing
toward that. But I can find out more on that.
SPEAKER: That's the inlet.
MR. MALLICK: Yes.
SPEAKER: Okay. I will ask. In looking at these
parameters, have you asked yourself are there any parameters
-- core inlet, I guess, doesn't do it, but core outlet might
-- any parameters that might be significantly changed as a
result of things like power upgrades and so on? Is there
anything in here, for example, that might be dependent on
flow rates?
MR. HACKETT: I'll try and take that one. This is
Ed Hackett. We haven't' gotten to that level of refinement,
Bob, but that's a good point. Among other things that
haven't really been considered here that may come into play
in the future, that would be one, power upgrades.
Another thing would be the change in the neutron
spectra relative to higher burn-up fuels or MOX fuel
possibly.
SPEAKER: It's really a shame. You play the game
with your hands tied behind your back and then somebody
comes along with an innovative idea and suddenly all of your
data is kind of -- it's not all that great anymore, and it's
your fault.
MR. HACKETT: It seems like that's what happens at
times.
SPEAKER: Just from the analysis that you've done,
how well do these sort of refined calculations match the
calculations that the plants used to estimate their
fluences?
MR. MALLICK: They are very much similar, I would
think so, but their details are not that - they have not
done calculations or it's not as refined. But there is not
that much difference, I would think so.
SPEAKER: Okay. So that even though you're doing
a more refined calculation, there's nothing to indicate that
the plant calculations are unreasonable or unconservative.
SPEAKER: But if they use a less dense grid, then
they get less peaking, don't they? I mean, their integrals
are the same or roughly the same. So you will show higher
peaks in general than they will.
MR. MALLICK: Probably so, yes.
SPEAKER: I guess -- I'm trying to think of the
right way to come back at that one. Bill posed the question
of which way would this go. I think this is a level of
refinement that's beyond what most folks would have
submitted, well beyond what most folks would have submitted
on the PTS rule.
And what they would have assumed there is look at
the maximum asmuthal fluence and assume that that applied
all around the belt line. That's what was historically done
before. Palisades was the first time, when we did the
Palisades PTS evaluation, this would have been vintage
'96-'97, that people -- that they first got into a
plant-specific fluence map.
And then what you're looking at is the integration
of that around the core and that always acts in their favor,
related to what they had done previously.
SPEAKER: Because you have a huge --
SPEAKER: Right.
SPEAKER: And everybody else just took that peak
all the way around.
SPEAKER: Exactly. Now, Bob is getting to the
point of how well that was modeled at the peak, and I guess
I don't have the wherewithal to come at that one without
Lambrose or somebody like that being here.
I think what was done is they would capture,
however the capture it, the peak asmuthal fluence and then
apply that fluence around the belt line.
SPEAKER: Okay.
SPEAKER: So they could tolerate a fair amount of
change and still be conservative, in all likelihood.
SPEAKER: I'm thinking about axial now, and that's
where all the structure is, or a lot of it.
SPEAKER: Good point.
MR. MALLICK: The next major activity that
integrates all the work together is the PFM code, which is
being revised and implemented. This objective is to
implement the refined PFM methodology as well as up-to-date
materials data into the code and make it consistent with
current PRA, as well as thermal hydraulics output data, as
well as methodology.
And myself, Nathan and Lea Berser, and Oak Ridge,
contractor, Oak Ridge National Lab, Terry Dixon, who is
integrating everything together, and University of Maryland
in terms of uncertainties and all those things will be
brought into this program.
Brief conclusion here, concluding remarks. The
analysis models are being finalized, such as embrittlement
correlation, fracture toughness distributions, and flaw
distributions. Then we are also going to -- based on these
finalized models, we're going to do some scoping studies
with reality doing some, but we are going to do a formal
scoping study on the particular plant, such as Oconee. The
application for the first plant at Oconee has started and
PRA, as well as thermal hydraulic area, but PFM analysis to
start soon on the scoping analysis.
But once we have finalized the whole model, actual
work on the complete analysis will start in the March
timeframe, we have modified other FAVOR code.
And just to comment, additional primary sources
are being used to build rigorous uncertainty model for the
key variables.
SPEAKER: Well, I congratulate the staff. Despite
the best efforts of the subcommittee, they've been right on
schedule.
We'll take a break now for lunch, and be ready to
start at 1:00.
[Recess.]
END TAPE 2, SIDE 2.
TAPE 3, SIDE A
SPEAKER: [In progress] -- the embrittlement trend
curves, and Mr. Kirk is going to give us the discussion.
DR. KRESS: Captain Kirk?
SPEAKER: Captain Kirk.
DR. SEALE: Shall we beam him up or beam him down?
MR. KIRK: That's why going into the Navy was
never an option, because I figured I might have some luck
with the career up to the level of captain.
DR. SEALE: Oh, there you go.
DR. KRESS: Then that would be it.
MR. KIRK: Then nobody would return your calls.
SPEAKER: With great foresight, he has put his
uncertainty analysis on the last view-graph.
DR. KRESS: That's a good idea. He knows what
he's doing.
MR. KIRK: Okay. I've got to reverse the order of
my slides, because I have the second presentation first.
SPEAKER: Well, the question is will we notice the
difference?
MR. KIRK: Well, I don't know. How much did you
eat for lunch?
Oh, here we go.
Okay. That works.
Okay.
The topic of the current presentation is revision
of the delta-T-30. That's the shift in the 30-foot-pound
sharpie transition temperature embrittlement trend curves.
My name is Mark Kirk. I work at Hackett's branch.
This information sees two applications.
One, of course, is the project that we're here to
talk about today and revision of the PTS screening criteria,
but the project that actually has generated the information
you're going to see here is another project that we're
working on on revision to Reg. Guide 1.99.
It will be Revision No. 3 when it finally comes
out, and of course, the application of that document is in
both a PTS assessment methodology where plant operators
calculate what their reference temperature for PTS is, it
then compares to the screening criteria, but it also gets
applied in the calculation of heat-up and cool-down curves.
So, in the development of this information, we had
those sort of dual applications in mind.
Now, the reg. guide itself will include
information and guidance on things that are not needed for
the PTS re-evaluation.
I've listed here sort of the -- this is the
high-level discretization of the reg. guide.
There is the transition shift embrittlement trend
curve.
There is the uncertainty analysis of that trend
curve.
There is the through-wall attenuation function,
because all of these -- all these transition shifts that
we'll be focusing mostly on here are calculated from
surveillance capsule data, and of course, that's bolted
right to the ID of the vessel. So, they're essentially at
ID fluence, ID spectrum.
That then needs to be attenuated through the wall,
so that's another thing going into the reg. guide.
We have treatment procedures for plant-specific
data and how we adjust for surveillance or not, and then we
also have upper-shelf energy trend curves and uncertainty
analysis.
Of those, these last two just don't come into the
PTS re-evaluation at all. All the other parts do.
The work to date and what the rest of the
presentation reflects is that the major focus has been up
here in getting the embrittlement trend curve, and that's
sort of where we are today.
The work is basically completed. We're in the
process of writing the technical basis document, and that's
an activity that's going to be going on among the NRC staff
for probably the next three to six months.
The embrittlement trend curve just became
available, or I should say the current manifestation of the
embrittlement trend curve.
The uncertainty analysis has just begun to be
performed, and the current view on that is that will be done
sometime in the November to December timeframe, although I
can share with you some early results from that.
We're just starting to have some discussions
regarding what the proper through-wall attenuation function
is, and similarly on treatment of plant-specific data,
although again, you know, just to give you a perspective on
this, the thought is -- and I'll discuss this in more detail
as we go through the presentation -- that probably as we
move to Rev. 3, we're going to be moving away from giving as
much credit to the plant-specific information and instead
going with more of the generic chemistry-based trends.
SPEAKER: When you say upper-shelf energy, is that
the JR curve?
MR. KIRK: No. Upper-shelf energy --
SPEAKER: -- means upper-shelf energy.
MR. KIRK: Yes.
SPEAKER: What about the JR curve work? Is that
going to be updated, the JR curve correlation?
MR. KIRK: Ed, help me out here. I wasn't aware
JR curves were in the reg. guide.
SPEAKER: They're not in the reg. guide, but
there's a JR correlation that has a through-wall
attenuation.
SPEAKER: It's a JR-curve-based attenuation, and
the answer is no plans for that right now, based on the fact
that it was the equivalent margins analyses that were done
with the industry to show that basically there wasn't a need
for it.
There are -- my understanding, although I haven't
paid attention to this a whole lot -- I think it was
addressed in Ernie and Bob's NUREG in 1998 in terms of a
refinement, but that refinement didn't indicate that there
was a need to re-do any of that work.
So, the short answer is no, we aren't going to be
pursuing that.
SPEAKER: Just to make sure I've got this right,
Ed, what goes in the reg. guide is an equation that predicts
the drop in upper-shelf energy.
SPEAKER: Right.
SPEAKER: Wouldn't have affect the
heat-up/cool-down analyses?
SPEAKER: It does, or it could. I guess I'd put
it that way.
The difference is, I guess, based on the
equivalent margins analyses, that it wasn't -- didn't look
like it was going to be any effect on plant safety for even
below -- significantly below 50-foot-pounds, which is where
the cut-off was in the 10 CFR 50, Appendix G.
SPEAKER: Okay.
DR. KRESS: You are going to share with us what
your perception of a plant PRA-consistent uncertainty
framework is.
MR. KIRK: Yes, sir.
DR. KRESS: Is that right?
SPEAKER: That's the last view-graph.
MR. KIRK: And I'll defer all the tricky questions
to Nathan on that one.
DR. KRESS: Okay.
MR. KIRK: Since I see he's sitting there smiling
at me.
Okay.
Just as a point of reference, the trend curve in
the current reg. guide that we currently regulation to is
shown here.
You've got your sharpie shift is a product of two
different factors, a chemistry factor and a fluence factor,
and absorbed into the chemistry factor are all the
dependencies of copper and nickel and product form. Those
are the ones that are explicitly called out in the table
that gives you the chemistry factor numbers.
When you see it in a few slides, the form of the
equation has increased considerably in complexity over the
years.
Where we started with this, to develop a new shift
curve, is that we've got considerably more data than we had
that Reg. Guide 1.99, Rev. 2 was based on.
Rev. 2 was based on something a little bit shy of
200 surveillance data points. We're now up almost to 800.
That's the database that Ernie and Joyce used to calibrate
the model.
The other thing that's changed considerably in the
past, I guess now, decade-and-a-half is our understanding of
the underlying physical causes of the embrittlement
mechanisms, and that has also played a role in the
correlation development.
So, the -- just a few notes on the modeling
considerations that were used in developing this
correlation:
It's -- for anybody that's looked at embrittlement
correlations, it's pretty obvious it's going to be a
non-linear fit, and as a consequence, some of the fit
coefficients are based on the entire data set, like, say,
the copper coefficients and the coefficients on nickel,
whereas some are based only on subsets, like there's a term
in there that expresses the influence of flux at low times,
and obviously, you can't -- or at long times, I'm sorry.
Obviously, you can't calibrate that with short-time data.
So, data subsets have been used in the fit.
Some of our metrics for what a good fit has is, of
course, minimum standard error, and Ernie and Joyce did a
lot of looking at the residuals.
Of course, they were looking for an average
residual, zero balance plus and minus residuals, but perhaps
the main focus in model development was looking at trends of
residuals where, of course, residual is just the difference
between what the model is predicting and what the original
data said, that there's no trend in the residuals with
either a modeled variable or an un-modeled variable.
If there's a trend with a modeled variable, then
that suggests you don't have the functional form right. If
there's a trend with an un-modeled variable, well, that's a
suggestion that perhaps you should include it in your model.
In terms of statistical significance tests that we
apply, our understanding coming from the physics and working
with folks like Bob Odette gives us some guidance in how we
run our statistical significance tests.
For example, if we have a variable like
phosphorous where we might not understand all the in's and
out's of phosphorous damage in a radiation environment, but
we do understand enough to say, well, if there is a
phosphorous effect, it's going to go in the positive
direction, that then suggests that you do a one-tailed test,
whereas if you have an element or an indicator variable or
whatever that you don't really know, then you'll be doing a
two-tailed test on statistical significance.
Also, the stability of the model was checked
extensively. Since it's a non-linear model, there's not
just one right answer, there's an infinity of potential
answers.
So, we check the stability of the fit coefficients
relative to the initial estimates by just making a bunch of
initial guesses and making sure that we always came out with
the same coefficients at the end, and also, we checked the
stability of the model relative to the data set used to do
the calibration.
The coefficients that actually came out were based
on a calibration of -- came from a calibration that used all
of the available data, but we wanted to make sure that the
trend curve wasn't over-fit and wasn't just somehow specific
to that data set.
So, we ran a number of calibrations on data
subsets to make sure that those coefficients came out
statistically similar to the coefficients in the equation,
where we used all the data, and indeed, they did.
Just to give you some examples of the type of
information that Ernie and Joyce were looking at as they
developed this correlation -- and I think these are graphs
you can probably better see on your hand-outs than in my
overheads, because the print's kind of small, but the upper
curves show the trend of both copper -- of shift with copper
and shift with phosphorous, whereas the lower curves show
the residual, the difference between what was predicted by
the model and these measure data relative to the final
model.
And of course, you see what we just said was the
criteria for having a successful model, that the residuals
do, indeed, show no significant trend with the modeled
variables, and indeed, if you look at the next slide, there
are, of course, other variables that don't appear, that you
won't find in the equation, like flux, specifically, and
manganese, but there were reasons to suspect that these
might be important factors.
Of course, the justification for leaving them out
is that you've got a model that has zero residual, a balance
residual anyway, so there's no burning need to put these in
at this time.
As we got -- I should note, this is an effort in
terms of -- when we go back in history, this is an effort
that probably dates back to about 1992, where we let a
contract with Modeling Computing Services to start looking
at developing this correlation.
They gave us a report in 1998, and we've been
doing some refinements on that model ever since, some of the
things, as we sort of came down to the 11th hour, some of
the variables that we were considering.
So, I should say -- I guess what I want to say is
there were other variables, of course, like copper and
nickel that were already in the correlation at this point,
but recently we've been looking at a copper saturation
effect, which you saw the empirical evidence of a couple
slides ago, that once you get to a certain amount of copper,
it no longer is damaging and the amount of shift saturates,
phosphorous, which you saw, and interaction between flux
time and fluence, which is to say that fluence isn't the
only descriptor of irradiation damage. So, we needed to
include other -- or potentially needed to include other
terms.
In looking at the data and in getting some new
data, there was also revealed what came to be called a
long-time effect, where the data points sort of at the end
of our statistical database, above 97,000 hours, show a
systematically higher shift than would be predicted by any
of the models that we had, systematically higher shift on
the order of 10 degrees Fahrenheit.
And then there was also an effect that was
discovered in the process of trying to find out what was
going here of vessel fabricator, where it was discovered
that, if you looked at the shifts in the plate data, those
plates that were in CE-manufactured vessels had shifts that
were systematically under-predicted by the model, whereas
plates in non-CE-manufactured vessels had shifts that were
systematically over-predicted by the model.
I'm not going to delude you that we have any
physical understanding, at least at this stage, of why the
heck that is, because -- I'll say it before anybody else
does -- 99 percent of those plates came from Lukens.
Now, of course, that's not to say -- there are
things that happen after the steel leaves the manufacturer's
shop and at the fabricator.
So, it's not completely implausible that something
like that could be true, but our physical understanding of
it right now is non-existence. There is, however, a very
compelling body of statistical evidence that the effect is
really there.
As we got down to these more -- what I will call
more nuancy effects than those of copper, nickel, and
fluence, we felt it was important to impose a bit of rigor
on ourselves in terms of thinking about, well, what of these
should we let into the model and what of these should we
leave on the table perhaps for next time. So, we developed
at least a gating criteria with a lot of fuzzy words in here
to help calibrate ourselves.
We said, well, we're trying to think about this
both in terms of what the statistical argument is for
inclusion or exclusion of a term, as well as what the
physical -- how well understood the underlying physics are
of the damage mechanism.
So, in terms of statistical basis, we looked at
the situation where we could have a strong statistical
basis, greater than 95-percent confidence that we have a
trend in the model that couldn't be attributed to a
mis-interpretation of random error.
You could have a weak amount of evidence or you
could have something in between, and then for physical
rationale, you could have a damage mechanism that's well
accepted, like copper, for example, all the way down to
something where you're sort of left scratching your head,
and like I said, this is -- you know, I drew lines in there,
but of course, in our minds, there weren't any hard lines
drawn, but certainly if you had something that was a
well-accepted rationale for the degradation mechanism and
strong statistical basis, well, of course, you'd include it.
If you had something that you couldn't see in the
database and you didn't know why, you'd never see it anyway,
but you would exclude it, and in between, you'd have to
exercise engineering judgement, but we tried to draw this up
to sort of guide our thinking.
Now as it turned out, when we actually ran the
statistics on the model, all the variables, or the effects,
I should say, that are being considered lately -- and by
lately, I mean within the last year -- came up very high in
the statistical significance category.
So, the physical rationale didn't enter much into
it.
I would like to focus, at least anecdotally on the
next few slides -- there's been some concerns expressed both
within the NRC and outside, within the industry. I should
warp ahead to say that our current proposal, so that we can
move forward and do the work that needs to come next, is
that we suggest to Terry for inclusion in FAVOR a model that
includes all of these terms.
The rationale for making that suggestion right now
is as follows:
There are certain things that you -- in order to
proceed, we need to have a model to proceed with. We can't
do an uncertainty analysis until we have a model. We can't
do a regulatory impact analysis until we have a model. We
can't do any sensitivity studies until we have a model
So, we felt it was important to suggest something
with the recognition that, in doing all of these analyses
and in further working on the technical basis document, we
may find things that make us say, well, no, maybe not, maybe
we don't want that in there.
But certainly, in recommending this model to Terry
in the PTS re-evaluation project, and as you can see by the
fuzzy words in our matrix, we did give definite deference to
statistical evidence over an existing physical rationale,
and like I said, there has been some exception taken to that
by both parties in the industry, as well as parties within
the NRC, and I just wanted to suggest that that's not a bad
engineering practice and, in fact, is fairly well-founded.
Sort of the essence of engineering discovery is
that we find out things by having field failures or by doing
experiments, and we might not understand the physical
rationale for why they're happening at all, but that never
stopped anybody from coming up with a design curve and
continuing to operate structures.
This just happens to one of my personal favorites:
In the 1860s, German railway axles were failing by
the truckload, and a gentleman named Wohler did a very
famous set of fatigue experiments where he developed what
was, in fact, the first SN curve, showed endurance limits,
and then those endurance limits were passed off to
designers, who then designed their axles to be below them.
Nevertheless, the physical understanding of the
phenomenon of fatigue at the time was wrong. There were
publications in esteemed scientific journals that said the
metal crystallized, and so, it broke.
That was obviously wrong, but it didn't stop the
design process.
Similarly, just another fun example, is that, in
1972, ASME developed an LEFM-based K1C curve that we have
used in vessel integrity calculations since that time and,
in fact, continue to use.
Nevertheless, the circa 1972 physically-motivated
prediction of the transition fracture phenomenon in foritic
steels did pretty well close to the lower shelf, but as you
got up off lower shelf, nobody understood the mechanism at
the time enough to predict this very sharp upswing that was
well-demonstrated by the data, but that didn't stop anybody
from believing the statistical evidence over the physical
model and moving on.
So, having now spoken heavily in favor of
empirical evidence, I should say that it is certainly not
the staff's intention to go only with empirical evidence.
Understanding the physics of what's going on is
especially important in this field, because we find
ourselves in the unfortunate but necessary position of
always having to extrapolate our data.
We never have data at the fluence or material
conditions that we actually are trying to predict. So, we
need to extrapolate all these trends, and that's why, in
what you'll see coming out of the technical basis document,
there is very definitely going to be a treatment of both the
physics of irradiation damage as well as the statistical
evidence of it.
In terms of the correlation, I thought I'd just
show the basic functional form.
It's got three terms in it, one related to stable
matrix damage, one to copper-rich precipitates, and then the
long-time bias, and you can see on the screen the various
input variables that go into each one.
The stable matrix damage is a function of
phosphorous, fluence, the product form, and the coolant
temperature, whereas the copper-rich precipitate term is a
function of copper, nickel, fluence, time, product form, and
manufacturer, obviously a more complex relationship than we
had previously.
We've done a few calculations, just sort of a
start of our regulatory impact assessment to see what
changes we might experience in going from Reg. Guide 1.99,
Rev. 2 to this new proposal, and like I said, this is very
early information, but I just show it for your information.
The graphs on the lefthand side of your screen
show the change in shift with the -- if we go to the
proposed model.
So, here, positive values mean that the new model
is predicting more shift than Reg. Guide 1.99, Rev. 2,
negative is less, divided it up into PWRs and BWRs,
obviously a lot of scatter between the two correlations, but
on average, for the PWRs, those that didn't have -- and this
all -- I'm sorry -- plotted versus the old Reg. Guide 1.99,
Rev. 2 shift.
For the PWRs, if there was -- if you had a
material that was low-shift already, on average, it might
get a little bit higher. If you had a lot of shift before,
on average, it's going to get a little bit lower.
The BWRs are higher by about 13 degrees Fahrenheit
across the board. So, the mean shifts are somewhat higher,
especially for the BWRs.
I've also summarized in this table a comparison of
the fit uncertainties, the new values coming out of the new
correlation work by Ernie and Joyce, and then the values
that are currently in the regulation, and you see that, for
the welds, the uncertainty seems to be going down a little
bit, but not a whole lot.
Now, in terms of your previous question on a
PRA-consistent uncertainty framework, this is where I'm
wishing I was able to do the second presentation first and
the first second, because I talk more about this in the
second, but we're developing the uncertainty framework here
using the same methodology as we've employed to characterize
the RTNDT K1C uncertainty, which is the topic of my next
presentation, regrettably.
The steps in the process are basically that we've
assembled the data and fit the curve, and that's what I've
been talking about, and that was done by Modeling Computing
Services and University of California, Santa Barbara, under
contract to the NRC.
We're now working at understanding the nature of
the uncertainties and developing a framework for a
mathematical model using a root cause diagram approach that
has been developed by Dr. Nitishan at Phoenix Engineering,
who is an EPRI contractor, and then that information is
passed on to Professors Maderas and Moseley at University of
Maryland, who term that sort of diagrammatic understanding
and physical understanding into a mathematical model that
then gets fed into favor.
That process will probably be a little bit more
well explained in my next presentation, but here is sort of
the -- again, the diagrammatic representation of what will
become a mathematical model in FAVOR, and I really don't
want to get into the details here, unless anybody wants to
drag me in, and then I guess I'll have to go, but what I
want to do is to point out a couple things.
This just shows the information flows from right
to left on the diagram.
The input variables are circled in yellow, and
this is how the math would actually be represented into
FAVOR. So, you'd have to know who the manufacturer was, is
it a weld or a forging, what's the phosphorous, the coolant
temperature, the nickel, the end-of-license fluence. You do
all that and then you can use the new embrittlement trend
curve to calculate a shift.
You compare that shift with information that you
might have from surveillance, decide if you're going to use
the shift or the surveillance data, and come out with a
predicted shift value.
So, the points to make on this is that, one, the
root cause diagram is, in fact, just an illustration of a
mathematical model, and that mathematical model allows
uncertainties to propagate through it from input variables
to output, and then the third -- and I also noted here that
this is the new embrittlement trend curve at node 14 here,
which, of course, has model uncertainty in it, and perhaps
the most important thing, from at least my understanding --
and I'm getting an education on this -- from a PRA
perspective is to distinguish between types of
uncertainties, namely aleatory and epistemic.
This is the diagram that helps us to understand
that. This shows the model uncertainty in the data that
Eason, Wright, and Odette used to develop the embrittlement
correlation.
So, for any given -- we can sort of work it
backwards, just so you can see.
For any given sharpie shift is, of course, just a
simple subtraction of a 30-foot-pound transition
temperature, un-irradiated, and a 30-foot-pound transition
temperature at some fluence, and that was determined from a
TANH fit to a plot of sharpie V-notch energy versus test
temperature, and then you can start to work it all backwards
and it then gets down -- the sharpie V-notch energy, of
course, gets down to very fundamental things like, well,
what was the chemistry, what was the heat treatment,
etcetera.
I should very quickly point out, this isn't a
model that ever gets mathematically run, but it helps us to
understand the natures of the uncertainties involved, and
Dr. Nitshan has color-coded it such that the epistemic
contributors to uncertainty are showed with the brown slash
marks, whereas the aleatory are shown with the solid brown
coloring.
And this is my interpretation, and this is, of
course, subject to more of the expert judgements of those
who know, but it sort of looks like the epistemic
uncertainty -- the epistemic contribution has to do with
things that are fairly well-controlled -- the test
temperature, the notch acuity, the machine calibration, the
test method, and so on.
So, while there is clearly, in any delta-T-30
value, components of both aleatory and epistemic
uncertainty, it would seem to me that the epistemic
contributors to uncertainty -- as an old lab rat, I'd say
these are fairly well controlled relative to some of the
things in the solid brown boxes.
So, one might come away from this understanding
with the conclusion that, while delta-T-30 does include, in
fact, both aleatory and epistemic components, perhaps it's
mainly aleatory, although that's -- you know, that's just a
poor man's interpretation of the diagram, but that's the
purpose of putting this together, and that's sort of a use
of this type of information, is to provide the materials
understanding to the PRA people and provide them with a
commentary that that they can understand to help make these
sort of decisions.
Just got a few slides left here.
Treatment of surveillance data: Currently, we
give credit for surveillance data in the form of a factor of
two reduction on the uncertainty and the shift provided the
surveillance data is deemed to be credible, and I don't
think I want to go into discussion of that, but let's just
say it's not always clear what credibility -- credibility,
in general terms, means that the data are well-behaved, that
you don't have something at 1E19th that's a shift of 200 and
2E19th that's a shift of only 50. That would not be a
credible data set.
But if you have at least two credible surveillance
points, by our current regulations one would be permitted to
reduce the uncertainty in the state of knowledge about the
shift by a factor of two.
There has never been any rigorous justification or
documentation of why that factor reduction is appropriate.
That's not to say -- I mean it's certainly completely
appropriate to update your date of knowledge based on
material or case-specific data. So, I don't ever want to
say anything that says, well, we're not going to do that,
because that's, in fact, the appropriate thing to do, but
our current plan is just not well-based, and at this time,
since that plan was developed, there's not been really any
work to give us a better plan.
The work that has been done has gone mostly on
what you just saw, into development of the mean curve.
So, the current proposal that's on the table is
that the -- sort of the default condition for the shift for
a particular plant or particular material condition will be
calculated based on the chemistry and all the variables in
the equation, you'll get a shift, you'll then compare the
predicted shift to a measured shift, if you have it, from
surveillance, and as long as it's -- you know, again, I'll
say reasonably close, as long as it's, say, within plus or
minus 2 sigma, one would use the shift predicted from the
model that's based on 800-some data points, rather than
adjusting that model-based shift to correspond to two
measured data points.
You know, again, that's the current proposal
that's on the table. That proposal is, I suspect, going to
be the subject of some fairly intense discussions among the
staff in the coming three to six months, but you've got to
start somewhere or you don't know what you're talking about.
An even perhaps more interesting one is the
through-wall attenuation question.
Right now, in Reg. Guide 1.99, Rev. 2., we figure
out what the fluence is at a particular thickness location
from the ID-X, where X is measured in inches from the ID by
taking the ID fluence and decaying it by this negative
exponential with the .24 coefficient.
Really, the only -- and I should say that there
hasn't been a whole lot of work since this equation was
developed that would give us a basis to do anything else.
There have been a very few test reactor studies where
basically a whole bunch of steel samples were machined,
blocked together, and then irradiated, so it simulated like
they were at different positions in the wall.
There was one study done like that which we'll be
looking at, and to my knowledge, that's -- I can't ever
pronounce it -- the Gundrumagin vessel. Those are really
the only data available that say anything to attenuation.
There's also the question of what the appropriate
damage -- radiation damage function is to use, should one be
using DPA or fluence to attenuate through the vessel wall.
So, there is some new information available that
the staff will be looking at. There's not a whole lot,
because quite frankly, it hasn't been an area of focus, but
there is a very practical impact that we need to look at in
that, with the old embrittlement trend curve, everything was
a function of fluence.
So, when you attenuated the fluence, you
attenuated the shift in direct proportion, according to this
relationship, whereas with the new equation, it's got terms
-- with the new equation, there are two terms. There's a
time term and there's a bias that don't depend on fluence at
all.
So, if you believe -- and now, this is a good
question, and as I said, again, I'm sure it will be the
focus of some interesting discussion over the next few
months. If you believe that this is really attenuating
fluence -- although when you look at Randall's basis
document, you decide that it might not really be attenuating
fluence. It's an engineering approximation.
Anyway, if you attenuate the fluence in the new
function according to that form and apply it only to the
fluence, then you certainly don't attenuate that and you
don't attenuate time, because time just marches boldly
forward.
Well, for some vessels, that's not going to matter
at all, because they're at such a high fluence anyway, the
contribution of that is nil, and so, you're giving up a
fractional degree, but for things like BWRs that tend to be
at lower fluence, it can be quite a significant impact.
It's going to be very important to -- like I said,
we don't have a lot of new information, but it's going to be
very important to consider the regulatory impact of this on
BWRs, especially for heat-up and cool-down, where we
attenuate the quarter-T and three-quarter-T to do our
calculations.
For PWRs and PTS calculations, the recommendation
that we've made to Terry right now is that, for right now,
pending further thought and information, use the Reg. Guide,
Rev. 2 function to attenuate the fluence in the new
embrittlement trend curve, it's not so much a problem in the
calculations that Terry is doing, because if a flaw is
deeper into the vessel than an eighth of a T, it's not going
to matter anyway.
So, this plot shows the impact of that
recommendation on the horizontal axis, is the old
attenuation at an eighth of a T.
So, this is how much less shift you had at an
eighth of a T than at the ID using Reg. Guide 1.99, Rev. 2.
This is how much less shift you have using the new
correlation but the old attenuation function, and what you
see is your have some situations, the heavy triangles or PTS
plants -- the worst it gets is you might have had one
material in one plant where it was previously attenuated by
15 degrees, now it's only attenuated by seven.
Again, the focus here -- I don't know if this has
come out earlier -- has been to get off the dime with the
calculations and get Terry something to use, with the
recognition that we might need to come back and change it
later.
It doesn't seem to be as significant an influence
here as it could be, certainly in this case. For the reg.
guide, it's going to be something that we have to very
carefully consider, because the impacts can be quite
incredible, up to 100 degrees Fahrenheit.
And I think that's basically it, just a discussion
of ongoing steps now.
SPEAKER: Doing that -- is that basically
equivalent to saying there's an aging effect that's
independent of irradiation?
MR. KIRK: Yes.
SPEAKER: And have we seen that -- you know,
except for when you dump a lot of philosophers into here, I
mean is there any --
MR. KIRK: That's a good question. Professor
Odette, in fact, just provided us with a report, which I
understand includes some of that information, but that, I
think, is one of the open questions that needs to be looked
at.
Of course, the difficulty being the availability
of material that's been cooked for that amount of time is
pretty low.
My understanding of what Bob's told me on the
phone -- and unfortunately, we just got the report last week
and I haven't had a chance to go through the details -- is
there is some information from hydro-cracker service of
similar materials at somewhat higher temperatures, but you
get into some pretty dicey cases of knowing when you're
extrapolating beyond the bounds of where you should be
extrapolating.
So, that's an open question, and in fact, of all
the terms in the equation -- and again, just a personal view
-- for my money -- well, this is probably the one that's got
people scratching their heads the most. It's like how the
hell did that happen? Bad luck.
I'm thinking that the procurement agent at CE was,
well, perhaps not as nice to the steel mill folks as they
were at the vendors, but that's just my theory.
But this is the one -- this term in here seems to
be the one that's the most theoretically contentious,
because some of the physical theories seem pretty good, but
getting the evidence to back them up in terms of what you
just pointed out, long-time data, is just very hard to do.
But as I pointed out by that graph, it has some
very significant practical implications in the heat-up and
cool-down mode.
SPEAKER: Now, I assume that Ernie has scrubbed
this looking for a phosphorous dependence, which would be
the -- you know, everybody's first --
MR. KIRK: I'm sorry. Scrubbed the long-time?
SPEAKER: Yeah.
MR. KIRK: That's a good question. I honestly
don't know for sure. That work sort of predated my
involvement. But that would certainly be a good question to
ask. I know he's scrubbed it every which way from Sunday,
but I can't swear to you that he's specifically looked at
that. You mean just looking for heavier incidents of tramp
elements in those.
SPEAKER: Right. Is there a reason an element
like that would be the prime candidate for just an aging
effect without irradiation?
MR. KIRK: Yeah. Of course, the feeling is that's
also showing up here.
SPEAKER: Right.
MR. KIRK: See, the thing is this got very
evolutionary.
This term came along first, in the historical
development of the model. There was the so-called flux time
term, and there was significant contention about that, and
we looked and looked and found more data, and then this one
popped up in trying to understand that.
This one actually exists independently of this,
because this is driven by data at low fluence, long-time,
BWRs, is what's driving the existence of this term. But in
collecting more data, we got -- of course, as time goes on,
you get more long-time data, and we found that, beyond
100,000 hours, the data points beyond 100,000 hours were
systematically under-predicted by the model on the order of
10 degrees Fahrenheit.
SPEAKER: These are very low fluences for kind of
an irradiation-assisted segregation, but --
MR. KIRK: True. Yeah, if you're looking for a
synergistic effect, you might want more atoms going through
it.
So, we found this looking for that, and then, in
saying, well, now, this really isn't making a lot of sense,
what's going on, this one popped up.
SPEAKER: That one's really tough.
MR. KIRK: But like I said, I can say to you with
confidence that -- I've worked enough with Ernie to know
that, if he says they're statistically significant, they
are, and the other thing that I perhaps should note is that
the industry group, EPRI and Sam Rizinski, contracted with
Dan Naman, who's a professor of statistics at Johns Hopkins
University, to take an independent look at this, and
Professor Naman came up with this quite independently of
Ernie, because we weren't letting Ernie talk at that time,
and Dan found it all by himself.
So, it's really there. I mean it frustrates
people, but it is, indeed, really there.
But in terms of where we're going on from here,
we're going the uncertainty analysis. Like I said, that
just got started, and we were able to turn over the
information to Dr. Natashan and Professor Medarez in August,
so they've sort of just started on that, probably looking
for seeing something out of that sometime in the November
timeframe.
We're working here on doing the regulatory impact
analysis and also having discussions about how surveillance
data should be treated, and of course, we're going to have
to get Ernie involved in those discussions, discussions
about through-wall attenuation, and we're in the process of
drafting the tech basis document for review.
And of course, the PTS project is going to be
continually updated, you know, on where we're going, and
we'll have to work with Shaw to see how that best slots in,
but what we did, I guess it was basically last month, is
gave Shaw and Terry our current best guess of, you know,
okay, if you hold a gun to our heads and say give me
something today, well, here you go, and what we've tried to
do is not just say here you go but tried to identify the
warts in it, so that nobody is misled that, well, you know,
this is true for all time. Well, it might not be.
But we also need to make very sure that we sync
the information that's in the reg. guide with the
information that's included in the PTS re-analysis, because
of course, we want both of those to be self-consistent and
supportive.
Any questions on that?
[No response.]
MR. KIRK: Okay.
Then my next -- should I go ahead?
SPEAKER: Yeah.
MR. KIRK: Okay.
The next set of slides is on fracture toughness
distributions and uncertainty analysis.
What I'd like to work through with you is talk
about our goal in doing this work and the folks that have
participated in what's become a fairly extensive cooperative
effort, talk about our approach, what new data we collected,
the uncertainty framework, show you some of the current
results, and then talk about where we're going next.
As I just suggested, there have been quite a few
people involved in this particular piece of work, and I
think to very good effect, because we've got a diversity of
experience and perspectives here that most projects have --
indeed, this is a fairly small-scale effort -- don't
normally enjoy.
I should say the goal here is to characterize
toughness for input into FAVOR in a way that's consistent
with current PRA methodologies, which is to say a proper
treatment of uncertainties.
At the NRC, I've sort of been coordinating this,
and Shaw and Nathan have both been involved.
At the University of Maryland, we've been working
with Professor Medarez to do the uncertainty work. His
graduate student is Faye Lee, and his associate is Allie
Moseley.
And at Oak Ridge, they've been involved in various
aspects of this work, both in collecting the K1C data and
developing a statistical curve, as well as more recently, in
looking at RTNDT model uncertainty. That includes Paul
Williams, Kenny Bowman, Terry Dixon, John Murkle, Richard
Bass, and Randy Nanstead.
And then we've also had significant support from
EPRI. Stan Rizinski is the project sponsor there, and he's
been kind enough to allow us to involve Marjorie Natashan of
Phoenix Engineering, and she's been doing the root cause
diagram work and interfacing directly with Professor
Medarez.
And what I'm presenting here is really the
amalgamated work of all those folks.
So, the goal I've already stated. The three boxes
below the goal show you the process that we've gone through.
We started off at Oak Ridge, and this probably goes over a
year ago now, assembling all the available valid K1C and K1A
data and developing a purely statistical fit to that. We
then moved on and involved University of Maryland and PAI to
establish sources of uncertainty using the root cause
diagram analysis to allow us to distinguish epistemic from
aleatory uncertainties and give us a procedure to treat both
parameter and model uncertainties, and then, coming out of
this, of course, our goal is a description of K1C and K1A
with uncertainties that we can plug into FAVOR.
This slide summarizes the work that was done by
Oak Ridge, now something over a year ago.
On the lefthand side, you have the K1C data;
righthand side, K1A.
The numbers in reverse video show you how the data
set size increased relative to that which was used to
establish the original K1C and K1A curve.
So, originally, we had 171. We wound up with 254.
The K1A database had a more substantial percentage increase.
The black specks on each of the diagrams is, of
course, the data itself.
The red curves on your screen are the way we used
to model the scatter in the data based on the ASME K1C curve
and moving that up and down by a sigma, whereas the black
curves are the new Oak Ridge model which is based on a Wible
formulation, and the same thing are shown over here.
One thing, just sort of looking at this and
saying, well, so what, that you come away with is you come
away with the immediate impression that the old scatter
bounds that we used in FAVOR were too narrow, especially for
K1A.
The consequence of that, the effect on the
calculated probability of vessel failure, of course, depends
upon the transients considered.
Terry did a nice little study of that probably a
little bit less than a year ago now, found in some cases it
mattered a whole lot, in some cases it doesn't matter quite
so much, depending on if you have late re-pressurizations,
whether arrest is important, and things like that.
Of course, this all needs to be considered in the
context of everything else that's going on.
The uncertainty analysis -- we started off with
the root cause analysis to identify the sources of
uncertainties, appealed to a physical model, so that we
could try to understand where the uncertainties came from
and distinguish -- Dr. Natashan and Professor Medarez worked
a lot together in terms of her trying to express the
physical understanding and the test lab understanding of
where these uncertainties come from to Professor Medarez,
who was working on the mathematical model.
So, they worked that out, developed a mathematical
model, which then Professor Medarez and Terry Dixon have
been working on to get it implemented into an actual FAVOR
programming structure.
The root cause diagrams -- and this is sort of
like talking about the horse after it got out of the barn,
because I've already done a couple of these.
It's just a way to diagram mathematical
relationship, show how uncertainties move from one place to
the next, but I do want to point out that the big change in
this way of doing things relative to the way we've coded
uncertainties in FAVOR before is that here we input
uncertainties and parameters back here and then propagate
those into uncertainties and output variables in a way
that's very systematic and critiqueable because you can see
it and say, no, that box doesn't belong here, it belongs
over there, rather than the margins being prescribed to the
analysis a priori, which is exactly what we used to do.
So, this is a very integrated and systematic
approach, and it actually works very nicely when you need to
get input from a whole lot of people in that you can lay it
out and explain it to them and they can see where -- how the
various pieces interact very easily.
So, it's worked out, actually, quite well.
Just to look at the diagram at its highest level,
for K1C RTNDT, of course, at the end, we want to get out the
uncertainty in K1C.
Going into that is the uncertainty bounds on the
fracture toughness data that I showed you previously coming
out of the statistical analysis of the data, but of course,
that's an index to an irradiated RTNDT value to position the
data in temperature space.
The RTNDT irradiated value itself is a function of
both an un-irradiated value and a shift, and of course,
there's a whole lot more that are in the detailed reports
but aren't shown here.
But again, similar to the last time, I do want to
make a couple of points here about some of the new or
significant features coming out of this analysis.
This diagram is just an expansion upon the one I
just showed you, and it even flows off on to other diagrams,
but one point is that we've got a process that matches or
models, I should say, the current regulatory framework for
how we determine RTNDT irradiated, and we're putting this
into the code, I should say, right for the first time. So,
that's a good thing.
We've got a statistical representation of
toughness that I already told you about, and that plugs in
here, and then, I guess the newest of the new things is a
recognition that there is a -- I shouldn't use the word
"systematic," because it's not always the same -- there's
always a bias in RTNDT. It's just simply not the right
value to use.
I can illustrate that to you -- and I should say
that's going to be taken account of in the calculations. I
can illustrate that to you just by putting up data from two
different heats of steel.
This is an A533B plate, HSST plate, are two tested
by Marsden back in '87 -- I'm sorry -- reported by Marsden,
tested long before that. This was the basis of the original
K1C curve.
So, you've got the K1C data and a K1C curve
indexed to RTNDT having absolutely no relationship to the
data other than the RTNDT value, was determined from
specimens cut from the same plate, and you see that, in this
case, RTNDT does a pretty good job of putting the curve
where you wanted it to go.
You can look at other data sets, like the
Midland-Beltline weld in the un-irradiated condition test by
McCabe in '94 and find out that RTNDT, in this case, is not
doing as good a job as you want it.
This is not at all unexpected. In fact, it is
expected, since RTNDT was designed to be a bounding, an
upper bounding estimate of fracture toughness transition
temperature.
So, we expect this to be the case, but it's highly
inconsistent with a PRA approach that's based on best
estimates.
We've got a parameter here that we use to figure
out where we go into our K1C distribution that we know is
always off, and it could be off anywhere from, say, zero
degrees to 150 degrees, and some accounting needs to be
taken of that.
Now, how that's going to be done, I can't tell
you, because we haven't quite figured that out yet, but will
it be accounted for, I think I can state unequivocally, yes.
We're still having some discussions between all of
the parties that I mentioned before regarding what the
correction function should be -- well, the correction
function being the probability distribution that relates
RTNDT to truth, however truth might be defined, and we're
arguing about what truth is, so I'm hesitant to put a
timeframe on that, and then -- that's perhaps a more sticky
question, and then we're also having some discussions,
mainly between Professor Medarez and the folks at Oak Ridge,
regarding what the proper mathematical procedure is to
create the correction once we know what truth is, and I'd be
the wrong one to talk about that.
But having said that, I think we're making
progress. We seem to be -- I think we're converging. But
we don't have an answer quite yet.
And Bill looks like he wants to ask a question.
SPEAKER: Well, I'm trying to figure out how I
know truth when I see it. What do I need to know to know
truth?
MR. KIRK: I could make a suggestion. I think --
this is going to really reveal my biases. I think the data
is truth. I mean you've got -- do you believe that linear
elastic fracture toughness characterizes the fracture
resistance of the material in an appropriate way for this
calculation?
If you can answer yes to that question, then you
say, okay, well, then truth is my K1C data for a particular
heat of steel, because that's what we need to characterize
to FAVOR, is the fracture toughness of the material on a
heat-by-heat basis.
So, then, if you can agree with those things, then
I think truth is some -- well, if truth is the data, then
the data is there, and the measure of how untrue RTNDT is is
just the distribution of -- for heat one of the steel, it
was off by 5 degrees; for heat two of the steel, it was off
by 100 degrees; for heat three of the steel, it was off by
50 degrees.
SPEAKER: Now, was this RTNDT determined from a
sharpie specimen of the same material as the K1C?
MR. KIRK: Yes.
SPEAKER: I'm not depending on a correlation to
get RTNDT.
MR. KIRK: No.
All these RTNDTs are, for what it's worth,
credible MB2331 RTNDTs determined on the same material, yes.
SPEAKER: Okay.
MR. KIRK: So, no, that's not in there making the
situation worse.
SPEAKER: Now, is it the same material seen under
the same flux, the fluence?
MR. KIRK: With very few exceptions, all of the
RTNDTs are on un-irradiated materials. There is only --
END OF TAPE 3, SIDE A; BEGIN TAPE 3, SIDE B
MR. KIRK: [In progress] -- four materials I think
we have a real RTNDT value on in an irradiated condition,
because quite frankly, most people don't irradiated NDT
specimens.
What I could tell you -- I don't have this graph
with me, but the -- really, what we're arguing over, the
distribution of the shifts, you know, what's the smallest
shift to what's the lowest shift, or how wrong can wrong be,
is always on the order of 125 to 150 degrees Fahrenheit.
What we're arguing over -- and this is where the
physics of it comes in -- what we're arguing over is where
this is positioned, but what I can share with you is that
we've made plots before of the various data points for all
the un-irradiated materials, and of course, we've got a much
larger number then that we do the irradiated.
The irradiated seemed to follow very closely to
the same trend, and I wouldn't -- I think the reason there's
a difference here is it's a test procedure problem, and it's
a stress state problem within the -- the differences in
stress state between the sharpies and the NDTs and the
fracture toughness.
The irradiation really isn't changing that
dynamic. I don't expect there to be a different
distribution. I can't demonstrate that to you very
convincingly with data, because I've only got four data
points, but I don't really expect there to be a difference.
But in answer to your question, I mean I think
here, you know, we've sort of -- well, in doing these
calculations, we premise truth on -- you know, that K1C is
an appropriate failure criteria, you know, for this material
under this application.
If we're going to question that, well, Pandora's
Box.
For me, I think the data is truth, and we need to
find out how far the RTNDT prediction is from the data. If
we don't believe the data, we've got bigger problems.
MR. HACKETT: This is Ed Hackett. I think I'd
just like to add sort of a tone commentary here.
A lot of this discussion sort of feels like we're
running down RTNDT, and of course, that was the basis of a
lot of good work that went into sections 3 and 11 of the
ASME code by a lot of folks who preceded us.
I think what Mark says is correct, but don't want
to leave anyone with the impression, because we've said we
don't maybe think it's a good indicator of the exact or more
accurate behavior of what we think we might see in a vessel.
However, it's worked pretty well for the ASME code in terms
of demonstrating in a convincing way safety assessments of
boilers and nuclear vessels and so on. Just don't want to
leave anyone with the impression that that's a problem, that
the current framework is a sound framework from that
perspective.
This would hopefully be an iteration on improving
the accuracy.
MR. KIRK: Yeah, and it works because it's been --
it's doing what it was designed to do. It was designed to
be conservative, and lo and behold, it is, you know, good
job, but that -- you know, again, my understanding, in
working with Mohammed and Nathan is that that doesn't really
fit very well into this approach, so we've got to do
something to try to take this -- you know, this being what
we understand it to be and what, in fact, it is and turn it
into a best estimate in our simulations, or at least correct
for the fact that it's not. I'm not sure if I'm saying that
quite the right way.
DR. KRESS: Well, you expect to do it on a
heat-by-heat basis?
MR. KIRK: Yeah, that's what you would be doing.
DR. KRESS: And then have a series of curves,
depending on the heat?
MR. KIRK: No.
DR. KRESS: You'd have one mean curve for all the
heats?
MR. KIRK: No, the idea would be, if you pick any
of these correction functions, just for purpose of
illustration, it doesn't matter which one, but you go
through the simulation and you decide, for a particular
region, I've got a particular set of material properties.
In a particular sub-region, that set of material properties
has a fluence associated with it.
DR. KRESS: Right.
MR. KIRK: I go through all the calculations and I
get a RTNDT irradiated, and then I go into a hat and I pick
up a number, and that's a number between zero and one, and
based on what that number is -- say it's .6, and let's say
I'm using this red one.
I would then take that number and reduce it by 80
degrees Fahrenheit, but I might go through the simulation
the next time, come up with exactly the same number here --
I'm sorry -- exactly the same estimate of irradiated RTNDT,
go into my hat, and this time pick up a .2 and decide that
I'm only going to reduce that number by 20.
What we're saying is this is our best state of
knowledge about how far off RTNDT could be, and since you
don't have any other information, you don't have the
fracture toughness data in this case, all you know is that
it's off by somewhere between this and that, and that, of
course, adds uncertainty to the analysis.
It also -- well, depending upon what -- it adds
uncertainty to the analysis, but it also adds a pretty hefty
mean shift.
DR. KRESS: Isn't that the same thing as drawing
the mean line through all the data and putting a
distribution around that line?
MR. KIRK: I'm afraid I don't understand what
you're saying.
DR. KRESS: That's all right.
MR. KIRK: Okay.
Just to summarize, what we've completed so far is
the statistical model of transition fracture toughness. At
Oak Ridge, they collected the data and made a fit to that
data.
In the development of PRA uncertainty framework,
we understood the current process that we use to calculate
an irradiated RTNDT using the root cause diagram approach
and develop mathematical models of that. Details of
implementing those models in FAVOR were discussed and
clarified between Mohammed and Terry, and ongoing work --
we're working on finalizing that mathematical model and
resolving the issue that we've just been talking about for
the past few minutes, the RTNDT bias, and also as ongoing
work, we're still working on assembling input data to run
all these models, and that's all I had prepared.
Are there questions?
SPEAKER: Does the uncertainty in RTNDT mean --
should you also go back -- the way you're accounting for
this uncertainty -- should that also have been included when
you did the fit to the K data?
MR. KIRK: Okay. That's one of the questions
we're considering under what the correction procedure is.
One proposal was the way I described it, which is to go
through, simulate an RTNDT irradiated in FAVOR, pick a
correction value, and get a corrected RTNDT.
Proposal 2, or 2(a) -- I've lost track -- is to do
exactly what you said, which is essentially to apply this to
the data and re-fit the data, take the consideration and the
uncertainty outside of FAVOR and allow it to be treated as
input data.
DR. KRESS: I think that's what I was saying.
MR. KIRK: Okay. I'm sorry. I didn't understand
it that way.
SPEAKER: It really puts it where it belongs,
because you don't know what RTNDT is when you're measuring
K.
MR. KIRK: True. And I think, Mohammed, would it
be true to say that's sort of the way the wind's blowing
now?
MR. MEDAREZ: I'm Mohammed Medarez from the
University of Maryland.
I think it's right. What we are trying to do now
is -- Oak Ridge is using a methodology, a bootstrap
methodology to shift the data by this correction, actually
the 256 or so data points that we have, and basically shift
it according to this curve that you have on the left.
There about about four or five ways of doing that,
and each of them have different implications.
We are in the process of doing that, and also, my
belief actually is that the two methodologies would yield
the same answer, although we are seeing some differences,
but I think we know what the differences are, why we are
getting those differences.
I agree, also, that it's cleaner to go and correct
the data, as opposed to, as you mentioned, actually go back
and calculate an RTNDT which is biased and then try to
correct it afterwards, but we have to understand exactly the
process here. We are not still there.
I think it will be about a month or two. Next
time, we should be able to propose a definitive process for
computing this error here.
MR. KIRK: Okay. Thank you.
DR. KRESS: You know, among everything else that's
here, I must say these materials guys have come up with the
sexiest slides produced in the last year-and-a-half.
SPEAKER: Break for 15 minutes.
[Recess.]
END OF TAPE 3, SIDE B; BEGIN TAPE 4, SIDE A
MR. DIXON: The title of this presentation is the
status of the FAVOR code development. I'm Terry Dixon, and
I'd like to acknowledge Richard Bass and Paul Williams, two
of my colleagues that work with me that helped me put this
presentation together, and the intent here of this
presentation is to describe the evolution of an advanced
computational tool for reactor pressure vessel integrity
evaluations, namely FAVOR, and basically, this presentation
is sort of broken up into five different categories.
The first one is going to talk about how FAVOR is
applied in the PTS reevaluation.
The second one is the integration of evolving
technology into FAVOR, the FAVOR structure, PRA methodology,
and the last one, which I'm sorry that Professor Apostolakis
left -- the very thing he was talking about, kind of
stepping through a calculation, was my intent here, assuming
that we have time.
Okay.
Someone asked this morning or alluded to this
morning, how would the results be used that comes out of
FAVOR, and this is an attempt to sort of answer that
question. So, application of FAVOR to this particular
effort, PTS reevaluation, addresses the following two
questions.
Here's a graph that shows frequency of vessel
failure as a function of effective full-power years.
Now, the abscissa here could just as easily be
RTNDT. It could be neutron fluence. In other words, you
could have this, different variables, but most people can
relate pretty well to effective full-power years.
But anyway, the two things that will be addressed:
at one time in the operating life does the frequency of
vessel failure exceed an acceptable value, which currently,
in the current regulations, is 5 times 10 to the minus 6.
However, someone presented this morning that this number is
probably going to change to 1 times 10 to the minus 6.
DR. KRESS: Look what that does to you on the
graph.
MR. DIXON: Yes. It could be dramatic.
Now, these curves, by the way, aren't -- these
don't correspond to a particular plant. This is just an
illustration.
DR. KRESS: But it could be a plant.
MR. DIXON: Obviously.
And then the second question is how does the
integration and application of the advanced technology
affect the calculated result, and by that, what we're
talking about here is -- say that, you know, you have a
model and you do the analysis, and at some time in the
operating life of the plant, say 32 years, shows that you --
that's how long you can operate your vessel and be in
compliance with the screening criteria to come back -- if
you improve your model, which is what we're trying to do --
this whole effort is to try to improve our computational
models, and you re-do it and you get a reduced value,
essentially what you've done is you have increased the time,
the period of time that you can operate your vessel and
still be in compliance with the screening criteria.
DR. KRESS: I presume, with that improved model
result, you make some guesses of what the changes would be
in the various parts of your model to get a different
result? I mean you kept everything the same, except you
looked at the -- for example, the K1C, you probably made it
less bounding and things like that?
MR. DIXON: We haven't done too much of this yet,
because as you've heard today, a lot of these models are
still being developed, but there was a paper that was
published.
DR. KRESS: We have a copy of that. I remember
it.
MR. DIXON: That was an attempt, as of about two
years, to do exactly what you're talking about.
DR. KRESS: Just to see if it's worthwhile to
continue.
MR. DIXON: Yes. It was like taking several
elements and saying, okay, if you change this to this,
what's the effect, and then what's the cumulative effect,
and that was sort of what kick-started this whole effort,
that that study showed that there was a potential, at that
time, for this type of -- in other words, to get additional
time in compliance.
DR. KRESS: If I were to look at the curve and it
was the only information I had and if I were to really
believe that the new acceptance criteria were going to be 1
times 10 to the minus 6, I might conclude that all this
effort is not worthwhile, if that were the acceptance
criteria, because you're not changing things much in a year
or two.
MR. DIXON: I couldn't say that until -- sometimes
you got to go down that road to know.
DR. KRESS: Yeah, you really do, I think.
MR. DIXON: You know, I couldn't make that
statement right now until we actually do this effort.
DR. KRESS: But it seems like it's pretty
important to pin down this acceptance criteria pretty early
in the game.
MR. DIXON: Right. But another thing that I would
like to point out here -- and it's referring back to the
question this morning.
Notice, this is just one line here. So, you can
think of this as being the mean value.
Now, every time that you execute the FAVOR code,
you get one point on this line.
In other words, you execute the FAVOR code at a
snapshot in the operating life of the vessel, in other words
corresponding to a particular fluence map that -- you know,
15 years, 30 years, 60 years, whatever.
So, you would run FAVOR at several times in the
life of the plant, and actually, you would get a
distribution.
Now, this doesn't show that, this just shows a
line, but actually, there is some uncertainty. We've
propagated the uncertainty through the model. So, this line
actually has a band around it.
DR. KRESS: So, my question earlier on was, once
you get that, what are you going to do with that?
MR. DIXON: Well, that's a good question, and I
don't know that we have that answer yet. I will just say
that that kind of gets into interpretation and regulation.
DR. KRESS: That's not your area.
MR. DIXON: Right. I'm not real sure that anybody
knows the answer to that just yet.
The schedule has been sort of sliding, but the
latest schedule decision is that, you know, the FAVOR code
will be ready for reevaluation analysis by around March 1 of
next year.
Now, in the meantime, models are being finalized.
You've heard discussion this morning about several of the
models. Then these finalized models have to be implemented
into the FAVOR code. Some of them are, some of them aren't,
and in the meantime, there's going to be scoping studies
performed specifically for Oconee, I believe it is, because
as Dave Besette said this morning, the Oconee thermal
hydraulics is essentially ready. I believe the PRA is close
to ready. We need the flaw data that was discussed.
So, all the input data, maybe not in a finalized
form but at least enough for us to kind of start cranking
some numbers.
Also, there was some discussion this morning about
kind of the history of FAVOR, how did it come about, and the
development was initiated in the early 1990s by combining
the best attributes of OCA and VISA with evolving
technology.
So, we show OCA-1, OCA-2, OCA-P -- all of these
were developed at ORNL in the early 1980s, and VISA was --
in the same timeframe, was developed primarily -- first at
the NRC and then later at PNL, and then there was lessons
learned from IPTS and a lot of lessons learned from the
Yankee Rowe experience, and Mike Mayfield was in Oak Ridge
at one time for a meeting, and I remember him making the
statement that the NRC was no longer going to support two
codes, VISA and OCA-P. He said I want a completely new
code, I want a new name, and I want it to combine the best
attributes of -- basically to do this.
So, that's what we've attempted to do.
There was public releases of FAVOR in 1994 and
1995, and then there was a limited release in 1999, a
limited release insofar as this group of NRC staff, industry
representatives and contractors, anybody that came to those
meetings got a copy of the code, and as I said, the current
development version is -- the plan right now is to be fixed
in March of next year for the PTS evaluation.
Now, of course, as you've seen, this is somewhat
dependent on other people feeding stuff into FAVOR. So,
this date is as good as the schedule that people feed things
in.
DR. KRESS: What kind of language is the code in?
MR. DIXON: FORTRAN.
DR. KRESS: Does it work on a PC?
MR. DIXON: Yeah.
Okay.
Kind of the second part of this presentation is
kind of the integration of evolving technology into FAVOR.
This is kind of schematic to show how elements of
updated technology are currently being integrated into the
FAVOR computer code to reexamine the current PTS
regulations, and this just shows kind of several blocks of
things that are done better now than they were done back in
the days of the IPTS and SECY-8265.
Detailed neutron fluence maps -- you've heard a
little about that. You'll hear a little more.
Flaw characterizations -- plates and welds --
you've heard a considerable amount about that.
Embrittlement -- new and better embrittlement
correlation that Mark Kirk talked about.
Thermal hydraulics -- the APEX experiments --
hopefully, RELAP, this latest version, confirmation through
experiments -- hopefully, we're getting better thermal
hydraulics data than we were 15 years ago.
PRA -- that's just kind of a generic term to talk
about kind of the overall methodology that I will talk about
in a moment.
RVID is the reactor vessel integrity database that
was created and is maintained by the Nuclear Regulatory
Commission that sort of, I guess, holds the official data
for every vessel.
If you wanted to know what the accepted chemistry
was for a particular weld or plate in a particular plant,
this is where you would go.
Extended K1C and K1A database -- the statistical
representations are -- I believe it's Professor Apostalakis
-- he said don't refer it to that way -- the uncertainty
representations of the K1C and K1A database. Again, Mark
talked about this. I'll talk a little bit more about it.
Fracture mechanics -- the FAVOR code itself -- in
other words, all of these are going to feed in what we would
say updated technology and we're going to apply this to the
four plants, which has been discussed, and then plot curves
like I showed a moment ago, and where are we, you know,
where are we when we do that?
DR. KRESS: Does FAVOR have the thermal hydraulics
built into it? Do you have to calculate the temperature
distribution through the wall?
MR. DIXON: Yeah. You're just one step ahead of
me, but FAVOR doesn't do thermal hydraulics. FAVOR accepts
thermal hydraulics as input.
In other words, output from RELAP becomes input to
FAVOR.
DR. KRESS: Yeah, but don't you have to translate
that into temperature distribution through the wall itself?
MR. DIXON: Yes.
DR. KRESS: But that doesn't come from RELAP.
MR. DIXON: No. RELAP gives you the coolant
temperature on the inside surface of the wall as a function
of time.
We'll talk about that, a little more detail about
that in just a moment.
DR. KRESS: Okay.
MR. DIXON: Okay.
This is getting a little bit redundant, I suppose,
but advanced technology is integrated into FAVOR to support
possible revision of PTS regulation, and again, this is just
sort of saying in words what we just said -- new flaw
characterizations, detailed fluence maps, improved
correlations, embrittlement correlations, reactor vessel
integrity database, better fracture toughness models.
Now, here is one that is very significant. FAVOR
will now be able to handle surface breaking as well as
embedded flaws, whereas previous versions of FAVOR, as well
as OCA, VISA did surface breaking flaws only, because all
the current regulations were derived from analysis that
assumed all flaws were on the inner surface.
Now, we include through-wall weld residual stress,
and then there's a lot to talk about in new methodology.
Certainly -- I referred -- Ed Hackett referred
this morning -- I referred already to the study we did in
1999 that showed a lot of potential existed for the
relaxation of the current PTS regulations, and the one
single thing -- Tom asked did we do sensitivities with
respect to different elements, and the answer is yes, and
the one that had the biggest impact was this significant
improvement in the flaw characterizations, when they
actually went and started cutting up -- non-destructive
examination as well as destructive examination of the P.V.
Roth, as well as Shoreham and other vessels, because the
current regulations, the current PTS screening criteria, as
well as Regulatory Guide 1.154, all your flaws are
surface-breaking flaws.
They took the Marshall distribution, even though
the data that Marshall distribution was derived from, the
flaws were, in fact, embedded, they said we'll still put
them on the inner surface. It was conservative.
But when they actually start cutting up the
specimen material, what they find is that there's a higher
number of flaws than what was postulated in the PFM analysis
from which the current regulations were derived.
However, all flaws detected so far are embedded.
In fact, Lee had some numbers up there this morning. When
you take the PV rough flaw densities and apply them to a
commercial PWR, you get about 3,500 flaws in the first
three-eighths thickness of the RPV vessel.
So, you're talking about considerably more flaws,
but none of them are on the surface, they're embedded, and
the impact of that was that you get considerably reduced
failure probabilities.
So, this, more than any single other thing,
element, showed the potential existed for impact in the
current regulations.
I pointed out lessons learned from IPTS and
lessons learned from Yankee Rowe.
One of them was that what we're dealing here with
-- we're dealing with an entire beltline, you know, and
typically we consider the beltline to be from one foot below
the core to one foot above the core, and the older codes,
OCA-P and VISA -- they would allow you to put, you know, one
chemistry, one neutron fluence.
So, you'd have to take kind of the worst case and
apply it everywhere.
But the current version of FAVOR now utilizes a
methodology that allows the beltline to be discretized in
the sub-regions, each with its own distinguishing
embrittlement-related parameters such as copper and nickel,
phosphorous, neutron fluence.
So, this accommodates the chemistries from RVID
and the detail neutron fluence map.
This just shows how, you know, you can break the
vessel up into different sub-regions, each with its own
embrittlement characteristics, each with its own number of
flaws, and so forth.
So, this was a pretty big step forward from the
older version codes to version codes that we have now, and
Brookhaven National Laboratory is generating very detailed
neutron fluence maps.
Shaw Malletshow talked about the number of points,
literally thousands. I mean they're talking about breaking
that vessel up into thousands of points, if you desire, and
this just shows some of the gradients.
Here's asmuthal location, and this is at mid-core.
So, this is 72 inches above the bottom of the core.
So, this is kind of the highest, and this shows
the asmuthal location at the mid-core, this shows it 13
inches above the bottom of the core, and this shows it, you
know, at the extreme top and bottom.
So, you see there's dramatic gradients here,
asmuthal gradients, as well as axial gradients. This shows,
as a function of your axial location, at core flats, and
this shows at some various other angular locations.
The point here is that there's dramatic gradients
and fluence that need to be accounted for.
DR. KRESS: That was a question I was going to
ask. Why do they need to be accounted for? Why don't you
just use the location that has your highest fluence and use
that as your -- that's where it's going to fail, right?
MR. DIXON: Well, I'll refer back to the figure
where I showed the two curves, where one is an improved
model. By discretizing -- it's guaranteed that when you
discretize and put in the map that includes these values, as
well as this, you're going to get smaller failure
probabilities. What you're talking about doing is doing a
bounding analysis, taking the highest value and applying it
everywhere.
DR. KRESS: That's because your flaws are
density-per-unit volume.
MR. DIXON: You will have just as many flaws here,
probably, as you do at this level.
DR. KRESS: Okay. How come the ones at that level
up there aren't the ones that fail, then?
MR. DIXON: They may be, but not necessarily.
SPEAKER: [Inaudible.]
DR. KRESS: That's the answer.
SPEAKER: [Inaudible.]
MR. DIXON: He said it better than probably
anything else I say. You've got to distribute everything.
DR. KRESS: That's the right answer, yeah. I
understand that.
MR. DIXON: Right.
So, think in terms of overlaying those fluence
maps, you know, those type fluence maps onto here, and you
know -- and you'll have flaws distributed over these
regions.
DR. KRESS: The question is what's the probability
of a flaw of given characteristics being at the same spot
that the high fluence is.
MR. DIXON: Right. By doing a Monte Carlo over
all of these permutations of possibilities, we feel you're
getting closer to reality.
SPEAKER: Why does that circ weld sit on that
axial plot?
MR. DIXON: Well, that would vary, I think, from
plant to plant, but I'm familiar with one -- I won't call
its name, but I believe the center line of this weld might
be about one foot above here, and a lot of plants, by the
way, will have an upper circ weld that falls into this
category. This actually corresponds to a plant, and I won't
call its name, but the whole idea here is you have one, two,
three intermediate axial welds, three lower axial welds, a
circ weld, you've got six plates, that when you went to the
RVID database, that's how much chemistry you would have.
So, I would call those major regions, you know.
This would have a different chemistry, and the
RVID -- it won't tell you that you had a chemistry here and
a chemistry here. It will just say this weld has a certain
chemistry.
The same with this plate and this plate, but it's
when you start overlaying the neutron fluence map onto those
chemistries that you get what I would call the embrittlement
map.
And again, why did we do this? Because when we
were doing -- when we were in the Yankee Row analysis,
evaluation and analysis, this was certainly a question.
People said, well, why can't you account for fluence
gradients? Well, the computational tools that we had at
that time just didn't. Nobody had taken the time to put
that in.
This is redundant. Mark Kirk's already discussed
this.
I'm just going to say that I'm talking about
things that are already into FAVOR code.
These new statistical models, statistical
representations, uncertainty models, whatever you want to
call them, for enhanced plane strain, static initiation and
arrest, fracture toughness, have been implemented into
FAVOR, and this just shows our 254 valid LEFM points, and
this shows the WIBLE distribution.
This is like the .001 percent curve, the 99.9999
percent curve, this is the median curve, and this very
lowest curve here is what, in the Wible distribution, is
called the location parameter.
There's three parameters -- A, B, and C. The
parameter A is the location parameter, and this is a plot of
that, and basically, that is the lowest possible predictive
value of K1C that you could ever have, okay?
Again, I guess if we were to get 10 more data
points, everything would change, but right now, that's where
we're at, and here it is for K1A.
Now, this was -- the old EPRI database, I believe,
was 171 points. We went to 254. I believe this was 50 or
54 data points. This one almost doubled. So, we've got
extended databases, and we've got much better uncertainty
representation of that data.
So, this is -- we feel this is a significant step
forward.
Okay.
Again, I've already alluded to this. This just
shows an inner surface breaking flaw, as opposed to an
embedded flaw, and as I mentioned earlier, the current PTS
regulations and reg. guides all deal with this guy, but what
is being found when they go out and do NDE and destructive
examination of vessel material -- they don't find these,
they find these.
So, if we want a better model, better
representation of what's out there, we have to be able to
model both inner surface breaking and/or embedded flaws.
So, the current version of FAVOR now -- it will
handle both, I mean at the same time. You can have a
combination of surface breaking and embedded.
Even though they haven't found any
surface-breaking flaws yet, it's my understanding that there
will be probably some surface-breaking flaws in the
characterization that goes into these analyses, because
perhaps they've looked at one-tenth of 1 percent of the
vessel material, and I don't think you would want to
conclude that, because you haven't found on there, doesn't
mean that there might not be one out there, and it becomes a
problem of statistics, and Lee Abramson is working on that.
Okay.
Just one slide here about the structure of FAVOR.
Maybe it helps. People that are code developers or code
users can relate to this.
When you talk about the FAVOR code, it's not like
just one module. It's actually broken down into three
modules, three completely separate modules.
The first one is what I'll call the load
generator, okay? And this top line of data is input data.
This middle yellow line -- that's the actual codes, the
executables. And this bottom line of data is output data
from each of the modules.
So, this module -- this first module is the load
generator, and the input to it is like the thermal elastic
material properties of the clad in base, the vessel
geometry, and the thermal hydraulic boundary conditions, or
in other words, the output from RELAP.
Now, the output from RELAP is going to be time
histories, coolant temperature, pressure, heat transfer
coefficient that's imposed on the inner surface of the
vessel, and FAVOR will allow you to give 1,000 pairs,
time-history pairs for each of those three for each of the
transients, and you can do 30 transients in one run of
FAVOR.
So, you can see this becomes a bookkeeping thing,
too. You're literally talking hundreds of thousands of
points.
Dave Bessette said this morning, for Oconnee, he
was going to give me 27 transients. Each one of those has
the three traces. So, 27 times 3 is 81, each one with
1,000.
We're talking 81,000 data points or time-history
pairs. So, we're talking about a lot of data.
DR. KRESS: Is that automated? You don't do that
by hand.
MR. DIXON: No. He gives me a disk.
DR. KRESS: He gives you the input curves?
MR. DIXON: He doesn't give me curves. I have to
make sure it's in the correct format, but that's relatively
simple.
But anyway, you feed this in -- and I'll talk a
little more about this in a minute -- you feed this in to --
basically, this is a finite element program, and out comes
your -- this is what you were asking, Tom, a moment ago.
You do get your space and time-dependent
temperature through the wall, how that gradient through the
wall at each location is changing as a function of time, the
same with regard to axial stress and hoop stress, and the
same for stress intensity factors for inner surface breaking
flaws, for different flaw geometries at different times.
Okay.
So, you run that module by itself. You run that,
and you get this output file, and it's just a lot of
numbers, but they're formatted in such a way that this
module, the PFM module, knows that format, and it can read
them in accordingly.
So, when you run the PFM module, you input the
flaw data, the beltline embrittlement data, all those
sub-regions and corresponding chemistries and fluence maps,
with all the flaw data, and also all of this load data for
each of the transients.
All that is used as input into the PFM module, and
out of that comes distributions for conditional probability
of initiation, conditional probability of failure for each
transient.
Now, it should be said that conditional
probability of initiation is dealing only with cleavage or
fast fracture. There is no EFPM. Somebody mentioned a
moment ago about JR curves. Okay. There is no ductile
tearing considerations going on.
This is a cleavage fracture, LEFM cleavage
fracture analysis only at this point.
Okay.
Then the third module is the post processor.
Actually, this only exists in my head right now, but I know
what to do, and the input to that is the transient
initiating frequency distributions, which comes from the PRA
people. Okay.
So, that's input, as well as these distributions
that you got from the PFM module. All that goes in the post
processor, and out of that comes the bottom line of an
analysis, and the bottom line is the frequency of
initiation. This is kind of a mismatch. It's the frequency
of RPV fracture, which is CPI, and the frequency of RPV
failure. So, the distribution -- that distribution would
have a mean value associated with it.
So, the mean value of this distribution would be
what was plotted in that figure I showed earlier, because
remember, I'm doing this at a moment in time, because
there's one fluence map here.
SPEAKER: Fracture in this case means initiation,
and failure means failure.
MR. DIXON: Good point. Initiation means fracture
occurs.
Now, whether that flaw propagates through the wall
is another question, and frankly, that's something we're
still working on, and I'll talk about that in a moment.
In fact, now we're going to shift gears and talk a
little bit kind of about, before we get too lost in the PFM,
probabilistic fracture mechanics detail, let's step back and
kind of talk about the overall PRA methodology.
This is a pretty busy slide, but this is just
showing that on the last -- on the slide a moment ago, I
showed the load generator. This is just showing the load
generator again.
But first, let me read this caption, because I
think this is important.
The FAVOR analyses incorporate the uncertainty
associated with the thermal hydraulics by including variants
for each of the transients, okay?
This shows RELAP generating a lot of output data,
okay, and major transients. Transient one might be a
small-break LOCA. Transient two might be a stuck turbine
bypass valve. Transient three, something else, dot, dot,
dot, transient N.
Okay.
Now, within each one of those major transients,
there's variance.
The way that a small-break LOCA comes down, it
could be this, could be this, could be this, could be this.
So, if you want to consider all those
possibilities, each one of these is three -- represents the
three time histories, each one of these errors. Maybe this
is a small-break LOCA, one possibility.
Here's the temperature, pressures, heat transfer
coefficient for that.
So, all of that goes in in one run of the load
generator, which performs a one-dimensional, axi-semetric,
finite element analysis to calculate the loads for each
transient, and again, this is redundant, temperature,
circumferential axial stresses, stress intensity factors,
tremendous amount of data here, big bookkeeping exercise
right here.
Okay.
The other module was the PFM module, and what it
does, it generates these arrays for the conditional
probability initiation -- I call that PFM-I -- and failure,
PFM-F, for vessel J subjected to transient I. It's starting
to get a little bit esoteric here, but think of this as
being a two-dimensional array, where each row in this array
corresponds to a particular transient -- in other words, one
of those representations that was shown on a previous slide,
and each column in this array corresponds to a vessel, and
the entry that goes into a particular I-J entry into that
array is the conditional probability of initiation that that
vessel fractured when subjected to that transient.
Same thing for failure, okay? And this module --
this is redundant. This is just another way of showing what
I showed a moment ago, where the loads, all the stresses,
temperatures, and everything that was done in the finite
element analysis is input into here, as well as the flaw
characterization files, which Lee and Debbie will provide,
for weld material, plate material; the PFM input, the
embrittlement maps for all those various sub-regions, along
with probabilistic input such as what's the one standard
deviation, you know, a lot of things like that.
I think I've about talked that one out.
A third module, if you recall, is a
post-processor, and the objective of the post-processor is
to integrate the uncertainties of the transient initiation
frequencies with the PFM-I and PFM arrays to generate
distributions for the frequencies of RPV fracture and RPV
failure.
This just shows the initiating frequency for
transient one, the distribution of initiating frequency for
transient one, two, dot, dot, N, okay? And these are shown
in histogram form, because it actually comes into the
program numerically. You don't say this is gaussian, this
is beta, because then you've got to create a whole library
of the possible distributions. So, we just said just do it
numerically.
So, that's the way that it's going to be done.
Also, the arrays that I showed a moment ago, the
PFM-I array, where the IJ entry, you remember, is the
conditional probability that that vessel will fail when
subjected to that transient, as well as the PFM-F comes into
here, and the output is the distribution of whichever one
you're doing, the initiation or the failure, okay?
So, you get a distribution, and this shows that
this distribution is, I guess, what statisticians like to
call bi-modal.
It will have -- typically, it will have a big
value, kind of a skyscraper here at zero, because hopefully
most of these were zero, and then you'll have some other
kind of distribution.
So, the mean of this distribution is not here.
It's going to be way over here.
So, it's going to be a very unsymmetrical
distribution.
Okay.
Now, the process to get here, what goes on in this
post-processor is that, for each vessel -- in other words,
for each column in one of those arrays, you sample the
initiating frequencies.
So, you have -- I like to think of it -- you'd
have a row vector of initiating frequencies, you know, one
value for each of the transients. Then you combine that
with like the PFM-I array, which I like to think of that as
a column vector.
So, if you multiply a row vector times a column
vector, you get a number, get a scaler.
So, that would be one value that would be the
frequency of initiation or, if you were doing failure, the
frequency of failure.
So, that would give you one value.
Well, if you do this, say, 100,000 times, you've
got 100,000 values.
So, you sort those, arrange them into a
distribution, then you calculate the mean and standard
deviation.
So, that's the bottom line right there.
In going back to that picture that I showed
earlier, you could take the mean of that, plot it for that
time, then you'd go do another analysis for another time in
the life of the vessel, and of course, what I didn't show --
and I'm being redundant -- what I didn't show on that first
slide was the amount of uncertainty, but we will know it.
Okay.
That could stop right there and that be the end of
the presentation, but we'll now try to talk a little bit
about some of the details of the PFM analysis. That was
just -- in other words, I'll try to talk about some of the
details of how you get a number into that PFM-I array, okay?
All I'm going to talk about here is how do you get
a number into the PFM-I array?
Now, I'm going to digress here just a moment,
because you asked a very good question this morning, Dr.
Shack, about -- you said you weren't sure if we were riding
one curve down or what, and I'm going to talk in more
detail, but now's a good time to interject that what I
showed a moment ago, in each IJ entry of that PFM-I array,
it's a number between zero and one.
Each entry has -- it's a probability, with the way
that we do it now.
The way we used to do it, which is what you said,
grab a curve, sample, ride it down, and either it's a
yes/no. It either breaks or it either doesn't. And that
was the old way of doing it.
In that case, it was a zero or a one. It's kind
of digital. It was either broke or it either wasn't broke.
But now, with our new methodology, you can have something
between a zero and a one.
Anyway, that will sort of lead in.
This is a terrible slide, and I'm going to maybe
try this a little differently. Instead of showing that --
that's even worse.
I'll try this, and like I said a moment ago, I
could have stopped, but we're going to jump off into some
details now.
The name of this section is PFM details.
Actually, I was hoping that it would be time for
people to go catch planes and stuff by the time I got to
here, but looks like not.
The idea here is -- remember, I said that I'm
talking about how you get a number into those arrays, okay?
I've showed you what you do with them after you get them.
How do you get a number? Okay.
I told you we're going to do many vessels. So,
let's let our outer loop be vessels, vessel equal vessel
plus one.
Then we know that all the vessels are going to
have multiple flaws. You saw Lee's presentation this
morning, and I had a slide here that showed that they have
around three or four thousand flaws, every vessel. So,
you're going to increment your vessels.
Okay.
Now, that particular flaw -- where on that
beltline region is it? Is it in a plate? Is it in a weld?
You choose that.
You sample and determine that. You place the flaw
on the beltline region, and in that beltline region, there's
a certain copper, nickel, phosphorous, neutron fluence, all
the embrittlement properties in there.
So, here, you've got a flaw located on the
beltline, with its embrittlement properties.
Now, we're going to sample the flaw
characteristics. How big is the flaw? Where is the flaw in
the wall? Is it a surface flaw? Where is it embedded?
So, now, we know enough to calculate the RTNDT of
the cracked tip. We know where the cracked tip is. We know
all the things that goes into the correlation that Mark was
showing, the chemistries, the neutron fluence. So, you get
an RTNDT.
So, at this point, we've got a flaw with a tip
located somewhere that's got a certain RTNDT.
Now, the next loop is going to be transients.
We're going to subject that to the various transients.
Okay. And the next loop is time, transient time.
So, we're going to step through here this time
loop, calculating the conditional probability of initiation
and failure for each one of these flaws.
SPEAKER: Let me ask my question here.
MR. DIXON: Okay.
SPEAKER: I've just calculated RTNDT. Why don't I
calculate a toughness?
MR. DIXON: Well, you do. You can see that this
was already pretty busy. This is high-level. That's the
next couple of slides of how we do that.
SPEAKER: Yes, but doesn't it make a difference
whether I compute the -- I pick that curve sort of outside
the time loop or I sample --
MR. DIXON: No.
SPEAKER: I guess this is my riding down versus --
MR. DIXON: No, either way. RTNDT is not a
function of transient time.
To calculate K1C, it's T -- it's a function of T
minus RTNDT.
T is transient time dependent.
So, I can calculate my RTNDT outside of even my
transient loop or my time loop, but you're right, once I get
into this time-loop, I'm going to be saying T minus RTNDT, T
minus RTNDT, and it doesn't matter if I'm moving down a
curve or moving across a distribution, my RTNDT is not going
to change.
It's the same RTNDT at that crack tip throughout
not only this transient but all the other transients, as
well, okay? And you're right, there's a lot going on in
here that I don't show, but there's some slides coming up in
a moment that attempts to address that.
But basically, you do this until all the time's
over, all the transients are over, you've done it for all
the flaws, okay, and then you have to go through this whole
multiple flaw thing. I'll talk a little bit about that. At
this point, you would have a value for one flaw, and then
you have to do kind of some algebra to combine the effects
of multiple flaws for that vessel, and we'll talk about that
in a moment.
And then the last -- you close your last loop,
which is vessel.
So, you set there, you got these four loops going
on, but physically, I like to think of it -- you know, you
take vessel one, you locate a flaw somewhere on that
beltline, you get an embrittlement, and then you set there
and hit that flaw with all the transients, okay, and then
you go to the next flaw, and you do that until all the flaws
are exhausted for that vessel, at which point you have an
entry into your PFM-I and PFM-F array.
I know that's a very busy slide, but it also
contains a lot -- what Dr. Apostolakis was asking this
morning. He would like to see you step through one
iteration. There it is. There's one iteration.
MR. HACKETT: Terry, let me add a comment while
you have that up there. This is Ed Hackett.
I think another thing that's come up in some
previous discussions with the committee is that it's
important to note that these are done -- as far as I
understand it, they're done randomly and independently. So,
there's no linkage, for instance, between an area that's
high in copper with some kind of idea that that would be
inherently more flawed than some other area.
Those are going to be, you know, in separate
loops, as much as something like that could exist. We're
not modeling that kind of thing.
DR. KRESS: But you did say you attempted to model
multiple flaws some way.
MR. DIXON: Yeah.
DR. KRESS: These are, you know, one flaw there by
itself.
MR. DIXON: Yeah.
DR. KRESS: And you're saying that there might be
another one close by and they link up or something like
that?
MR. DIXON: No. Yes, but right now, let me just
say -- maybe I'll just say this. Right now, there is no --
right now, there's an assumption that every flaw is
independent from every other flaw as far as fracture. The
presence of one flaw does not influence the fracture
response of another flaw.
However, at the PVP conference in Seattle this
past July, a professor from the university of Ottawa
presented a paper that I went to, and he had done some work.
So, I think -- I've read his paper.
I actually think -- I don't know if we want to,
but I was going to discuss it with NRC staff at some point
in the future.
He's got curves that you could use to sample that,
but I'm not sure that we want to go there. I don't know.
His work was kind of the first, I think, in this area.
Right now, the answer to your question is every
flaw is independent of every other flaw.
SPEAKER: How long does it take for a single run
from vessel equal vessel, from the first vessel that's
chosen to the end point?
MR. DIXON: Okay. That's a good question. It
depends on a lot of things.
I've got a machine that's a 533-megahertz machine,
and to run it for, say, 100,000 vessels, for a single
transient, 100,000 vessels, where each vessel has around
3,500 flaws, it's about -- like I'll start it when I leave
work, like at five o'clock, and I'll come back the next
morning and I'll see where it finished at 2:30 in the
morning or something.
So, it's eight, nine hours on a 533-megahertz
machine for one transient, and Bessette said this morning
that he's going to give me 27 transients for Oconee.
So, I can already see that we may have to -- I
know, right now, you can buy 800-megahertz machines for the
same thing that you could buy this one for last February.
So, I think we may have to -- maybe by next March,
when we get ready to do this, we may go buy us a couple
giga-flop machines, which will probably be out there for
what we bought the 533 for last year.
So, I mean you can see that this is pretty
computationally intensive.
And remember, at the end of the day, when you do
that, that's just one point on your curve.
Okay.
I told you that I would try to -- between that
transient time loop -- I just stepped over it. Now I'm
going to try to address that a little bit here.
Here's a transient. In fact, this is taken from
the IPTS studies.
This is designated in the IPTS studies as Calvert
Cliffs transient 8.3, and it has a distinguishing
characteristic that was a distinguishing characteristic of
most of what was called the dominant transients in the IPTS,
those that contributed most significantly to the vessel
failure, and that is this late re-pressurization.
You know, your temperature is here. It's a pretty
sudden cool-down down to about 150. No, it's not very
sudden. It's pretty gradual. Over a period of two hours,
it cools from 5.10, I believe, to around 150.
Pressure drops suddenly, stays low. Get over here
about 95 minutes, boom, you spike back up to full pressure.
Bad news transient.
But anyway, I'm going to use this transient to
illustrate this new methodology of calculating the
conditional probability of initiation, as opposed to the old
way of going up and getting a curve, picking a curve and
riding it down, and either the vessel breaks or it either
doesn't.
Okay.
This is a lot of words. I'll read it.
The conditional probability of initiation is
calculated by solving the Wible K1C cumulative distribution
function for the fractional part, percentile, of the
distribution that corresponds to the applied K1 as a
function of T, a lot of words, but what that means -- what
this is an attempt to illustrate is here's your Wible
location parameter. I showed earlier, that's the lowest
value of K1C you could every have, okay? And I chosen an
arbitrary flaw. I said let me take a half-inch-deep flaw
that's embedded, that's located such that it's inner cracked
tip is one-half-inch away from the RPV inner surface.
So, I've got a flaw that's a half-inch,
through-wall, located a half-inch from the inner surface of
the vessel, and I subject that to this transient, and here
is the K1.
Now, this is T minus RTNDT. So, time is going
this way, okay?
This shows the applied K1, this K1 as a function
of T, moving this way, and you notice that it never breaks
into the -- it never penetrates the K1C space until the
re-pressurization.
At 95 minutes, about 95 minutes, boom, it spikes
up here, and at that point, that is the 6.35-percent curve,
okay, or the .0635, which you solve if you put the K1 into
the Wible cumulative distribution function along with A, B,
and C, which are functions of T minus RTNDT, you get the
conditional probability of initiation for this transient at
this time for this flaw, okay?
So, this is pretty fundamental right here of
what's happening down at the innermost kernel of this
algorithm, okay?
Now, here's another -- here's an attempt to show
that same thing another way.
In the illustrative example problem, the Calvert
Cliffs, 8.3, at the time of re-pressurization, K1 is greater
than .0635 of the Wible distribution at this particular
vertical T minus RTNDT.
So, at that moment in time, when you spike up
below that lowest value, the question is how far did you get
up into that K1C space, which I showed how you solve for
that, but all you're doing is you're just solving for what
part of that total distribution is applied K1 greater than,
okay?
Now, if you want to ask questions, this is a good
time to do it, because this is new. This is new PFM
methodology that -- basically working with the University of
Maryland, and it's my understanding that this includes the
aleatory uncertainty that we used to didn't include.
When we used to get up and ride a curve all the
way down and it was either a zero or a one, that did not
include the aleatory uncertainty, whereas this method does.
SPEAKER: But it says that that variation in K is
all aleatory.
MR. DIXON: There's no variation in K. K is only
as a function of time.
SPEAKER: K1C.
MR. DIXON: Right.
SPEAKER: Somehow, I would pick that as families
of curves for a given material.
MR. DIXON: It is families of curves. It is, in
fact, families of curves. You can think of it that way.
SPEAKER: But once I've picked the material, I
have a curve, with perhaps some scatter around it.
MR. DIXON: You're right. Once you pick RTNDT,
you have -- I'll tell you what. Maybe this will help.
Maybe it won't. We can go back to that slide.
This is an attempt to show -- this is showing it
as a function of time. Now, you know, we're moving this
way. This is a different situation. This is not that
transient 8.3. This is a different case, a different flaw.
But this shows how the Wible location parameter changes as a
function of RTNDT. As RTNDT increased, that Wible location
parameter gets lower.
Now, here comes this -- in time, here comes the
applied K1 in time.
So, the question is, how much does this applied
K1, if at all, how much does it penetrate the K1C space, you
know? That's the question that we're asking when we do this
particular computation, and the little dots correspond to
the discrete times that we're analyzing it at.
Now, this is a plot of the instantaneous
conditional probability of initiation; in other words,
solving -- as I showed a moment ago, solving the Wible
cumulative distribution function as a function of time, or
in other words, as a function of applied K1.
You can see that, at 325 degrees, RTNDT, which is
pretty high -- I did it just for a good example -- how far
is it above this line, and for 275, how far is it above this
line, and this answers those questions.
This shows the conditional probability of
initiation as a function of time.
I don't know if that helps or not.
SPEAKER: Let me just take a more simple-minded
approach.
MR. DIXON: Okay.
SPEAKER: If I went back and, you know, I plotted
that data, all my 274 data points --
MR. DIXON: Yeah.
SPEAKER: -- for the K1C, and I have all the data
for a single material, you know, where I've made all the
samples, will those sort of occur randomly within that band,
or will the material for a given sit somewhere either at the
top, bottom, or middle of that band?
MR. DIXON: I don't know. Mark could probably
answer that better, and he stepped out.
SPEAKER: Could you give that one more time, Bill?
I'm not sure I followed that.
SPEAKER: If I take my 254 data points, and those
are multiple heats of material, and I look at a single heat
of material, will I find it uniformly scattered up and down
that band, or if I look at single heat of material, will I
find it sitting somewhere in the middle of that data as I
move from RTNDT?
SPEAKER: Looking at a single heat, I'd be
inclined -- I guess I can't answer for the current
situation. I think previously I know -- I can say the way
we addressed Pallisades, it would have been uniform, is that
way we've done it previously, and I don't know if that
carries through to where Terry is now.
SPEAKER: Yeah. He's saying you can go anywhere
from the top to the bottom.
SPEAKER: Then that's a random choice.
SPEAKER: That's a random choice, whereas, you
know, I'm sort of -- I would have argued that maybe that
band really indicated that some materials are tougher than
others, and therefore, you pick a material and you would
have had some aleatory distribution, but it would have been
a much narrower aleatory distribution.
SPEAKER: I see what you're saying now. I think I
understand now.
That would be the intent of the new methodology,
would be what you just said there.
SPEAKER: No, I think the new methodology says I
go up and down the whole damn curve.
SPEAKER: Yeah.
SPEAKER: See, I thought it was more the -- you
know, this is going to depend on how these uncertainties,
you know, cascade into this, but I would have thought it
would be more what you just said. Maybe I've got the wrong
impression.
MR. DIXON: Going back to our K1C database --
END OF TAPE 4, SIDE A; BEGIN TAPE 4, SIDE B
MR. DIXON: [In progress] -- a vertical slide
through there at a given value of T minus RTNDT.
Now, I don't know if what I'm fixing to say
addresses your question. I may not get this exactly right,
but you'll get the idea.
That 254 data points -- I believe there was 16
groups, okay, 16 groupings of various T minus RTNDT, okay,
you know, plate, HSST, one four plate, HSST, 02, dot, dot,
dot, and so, they were grouped by heat, but the Wible
distribution that is derived from that does not include
those considerations. It's just data.
SPEAKER: Right. That's okay if the data for all
those plates sort of falls up and down that thing uniformly,
but if they were all colored and I saw all green balls down
at the bottom and I saw all red balls up at the top, then
doing my Wible -- I can't answer my question until I know
where the balls lie.
DR. KRESS: Are you going to be able to know which
heat a given vessel --
SPEAKER: No, but then I would sample -- I don't
know where the curve is, and so, I would sample -- you know,
what I would think of as families of curves and pick a
curve.
DR. KRESS: Yeah, but on what basis would you pick
that curve?
SPEAKER: Because it would be some material, and I
would pick it at random, but once I picked that, I would say
-- the material never changes through the whole transient.
Every time he goes to a time step, he goes up and down that
whole distribution, and I would say no, once I've picked my
material, I've sort of got a tough material --
MR. DIXON: No, no, no. What you just said is not
correct.
SPEAKER: It's not?
MR. DIXON: What you just said is misleading.
Keep that picture in your mind.
Now, let me see. Let's go here.
We're not bouncing up and down. The question is,
the way I like to think of it -- I don't know which picture
is best.
We're not bouncing up and down anywhere. The
question at any time is what percentage of the Wible
distribution is the K1 greater than? That's not bouncing up
and down.
In other words, if I was to -- in fact, one of my
back-up slides may do this. No, it will just confuse it.
In other words, if you were to go at this time, 10
minutes later in the transient, you don't come over here and
completely re-sample. You're not sampling. There's no
sampling going on here.
The question is, you've got this K1C space
defined, between here and here.
Now, the question is, when I bring my K1 as a
function of T into play, how does it penetrate that space,
if at all? That's the question.
There's no sampling involved.
DR. KRESS: But you're saying if you define that
curve a little finer, with the different colors, you could
have sampled it.
SPEAKER: If I can interject here, I want to point
out a couple of things.
One, the curves that Terry is showing here are
certainly necessary for us getting on with the work and
formulate things, but I don't think this is the final word.
This was based on the statistical analysis of the data set.
If you go back to Terry's slide showing the Wible
equation, he mentioned the parameters A, B, and C. One can
develop different distributions for those parameters.
That's where the epistemic comes in, what's the value of A,
B, and C, and if you segregate the data based on different
characteristics -- and I'm way beyond my depth now, but
conceptually, what you would do is, if you identified
different classes for which you have different families of
values for A, B, and C, that's how you would enter that
process.
So, that would be the same thing as what you're
talking about, selecting the curve. In this case, you'd be
selecting A, B, and C, and then, once you have that now --
SPEAKER: It depends on whether you're doing that
inside the delta time group or outside the delta time group.
SPEAKER: That's right, and the epistemic loop is
outside, by definition. The inside is when you're dealing
with the aleatory component, because now you're dealing with
a transient and the response on a time step by time step
basis.
SPEAKER: When I say bouncing every time, Terry,
at every delta-T, you're sampling a K1C.
MR. DIXON: No.
SPEAKER: Where do you determine the K1C? That's
outside the delta time group?
MR. DIXON: There is no sampling of K1C. Once
you've got your -- you've got your K1C space defined by the
Wible statistical representation.
Now, the question is -- I'm going to put the K1
into that function, and what I get out of that is the
percentile K1C curve or which one of those family of curves,
if you wish to look at that way, does that K1 correspond to?
Let me try it this way.
This is the Wible cumulative distribution function
that, if you had K1C in here, if you had K1C in here, it
would tell you which one of those families -- in other
words, which percentile K1C curve is that, as a function of
K1C, A, B, and C, but when I plug K1 in instead of K1C, the
question that I'm answering is what -- I mean, right there,
that shows -- that's the 6.35-percent K1C curve.
So, I'm not sampling K1C. I'm asking the
question, how far does my K1 penetrate into K1C space?
MR. KIRK: Mark Kirk, NRC.
Can I say it maybe a different way, relative to
Terry's ugly slide? Maybe we've found a use for that.
Could you put it back up?
MR. DIXON: Okay.
MR. KIRK: Terry's certainly right in what he's
saying, he's not sampling K1C, but the material properties
for any given -- if you look at the loop that says calculate
RTNDT at cracked tip, that's outside the time-loop.
So, at that point, I guess the way I'd think of
it, once he's calculated RTNDT at the cracked tip, at that
point, he's determined where the K1C curve is for that
material. That's then fixed on a toughness versus
temperature plot.
He then goes and runs the time loop, and that's
what the illustration -- if you go to your slide 23 -- so,
once he's determined RTNDT at the crack tip, he's determined
where the K1C curve is for the whole time loop.
He then runs the time loop, I think, as he said,
from right to left, and that's the applied K1 changing with
time, and how it winds up with in the K1C curve gives you
your final probability of failure -- of initiation, I'm
sorry. But once you get inside the time loop, the material
characterization has been fixed. It's not re-evaluated each
and every time.
SPEAKER: I see what he's doing now.
MR. KIRK: Is that a correct interpretation,
Terry?
MR. DIXON: Yeah.
Notice, at these points down here, you can
positively say the conditional probability of initiation is
zero. It does not get equal to or above this lowest
possible value of K1C, the location parameter.
You can positively say, you know, with a
confidence interval very high, that the probability here is
zero, until you re-pressurize.
SPEAKER: [Inaudible.]
MR. DIXON: Yes. In other words, any time the K1
is above this location parameter, you've got a non-zero
value of conditional probability of initiation.
SPEAKER: Could you put that other slide back up,
Terry, the schematic again? I just want to see if I'm clear
on where Bill was coming from.
MR. DIXON: This one?
SPEAKER: No, the methodology.
MR. DIXON: Oh.
SPEAKER: Because I was wondering -- maybe I'll
pose it as a question, Bill.
I was wondering if you were on the third box down
where we're looking at sampling sub-regions and where that
relates to the generation of the K values in terms of maybe
compartmentalizing the K1C or that type of generation,
because obviously, we are looking at different values for
the different sub-regions, but that also, by Terry's chart
here, is fixed outside the loop, outside the transient loop.
I don't know if that helps at all, but I was
wondering if that might be where you were coming from.
MR. DIXON: The only material property that's
varying in here is K1C, and it's varying because temperature
is varying. RTNDT is fixed.
In this loop right here, your temperature is
changing. Therefore, T minus RTNDT is changing.
Mohammed?
MR. MEDAREZ: Mohammed Medarez. Maybe if I can
show you --
MR. DIXON: Sure.
MR. MEDAREZ: This one view-graph -- maybe it
explains this a little bit better.
If you're looking at it, here's the K1C
distribution, and as time goes by, the distribution will
have different shapes.
It slightly changes, because as time goes by, the
temperature changes slightly.
Typically, if you take a sample of this as a
percentile here and if this is your K1, this shows the time
that it exceeds, okay?
If I take many of these samples, I can build a
distribution here of the time that I initiate that flaw.
Everything inside a flaw, flaw is fixed, and I'm just going
over time now, okay?
So, I take -- typically, I think this is what you
do. You take a sample here, and this is a sample of the
percentile. From the old time, he had only a bounding
value. Now he has a distribution of these, because he has a
variability, and therefore, he gets a distribution of the
time to initiation of the flaw.
So, for instance, the probability that he would
have any initiation between this time and this time in this
area which is hatched, which is also equal to that area.
So, that's the difference from the last time of
operation.
Essentially, he used a bounding line, and now he
is taking a percentile of this curve, but he stays constant.
Once he takes that, he stays constant over that line, and
finds what time the crack starts to initiate.
So that's the process.
SPEAKER: Why is your cumulative probability on
the bottom -- why doesn't that go out to where your K1T
crosses your bottom line again?
MR. MEDAREZ: This one, why it goes down?
SPEAKER: Your probability that you're
accumulating the probability of failure.
MR. MEDAREZ: Because physically, if you start in
here, you started right here, if it goes down, it has
already started. So, you don't start it again. That's why.
If it goes up and down, it can only start one
time, and that's it.
So, that's why you have, actually -- once you
reach the maximum, you trap it out completely. There is
nothing else after that.
MR. DIXON: I don't know if this will help, but
Mohammed basically is saying, okay, given this applied K1 as
a function of time, you could set here and do a Monte Carlo
analysis on this flaw and sample this Wible K1C distribution
and come down here and get a distribution.
What I'm saying -- and we have verified this, he
at the University of Maryland, as well as I at Oak Ridge --
you get exactly the same answer as if you go ahead and
algebraically solve the cumulative distribution function.
It's the same thing, because if you do this Monte Carlo,
which becomes computationally prohibitive, because now
you're doing a Monte Carlo within a Monte Carlo, and that
gets a little bit crazy, but what you're really asking is,
you know, what's the percentile of your K1 space that you
penetrated? That's the way it comes easiest for me to
understand.
SPEAKER: I don't think we should get too hung up
on the -- I mean there is a difference between the mechanism
used to do the computation, and we can use sampling or we
can use quadrature, we can do lots of things, and then the
basic model as to where the variability is coming in, as a
function of time, and where the epistemic uncertainty and
how that's treated, and clearly, we need to do a better job
of explaining that.
So, I think, in the upcoming meeting, we will
certainly put together a better story as to how your issue
is being addressed, because we understand the question.
SPEAKER: Okay.
MR. MEDAREZ: And right now, of course, we're
treating it as aleatory, but we recognize that that may not
be correct.
SPEAKER: There are components that are epistemic.
You're not seeing that right now in this curve.
MR. MEDAREZ: But right now, we basically carry
the whole uncertainty through, and what we're calculating is
the probability of vessel failure, which is all aleatory, in
that case.
SPEAKER: I guess what I missed was the fact that
you're really looking at these cumulative curves.
MR. DIXON: Shaw, did you tell me that you
distributed to these guys a copy of that IAEA paper that we
wrote?
MR. MEDAREZ: Yes.
MR. DIXON: It's called updated probabilistic
something. It's a paper Shaw and I wrote for the IAEA
conference.
That says in words what I'm getting tongue-tied
trying to say up here. In other words, that problem with
that re-pressurization -- there's a narrative that describes
that in that paper that probably says it better than I'm
trying to say up here right now.
I can write better than I can speak.
SPEAKER: I'm not sure I completely understand
everything, but now I understand what you're doing.
MR. DIXON: Until now. And this is a very
complicated looking slide, but -- and I probably made this
more complicated than I had to.
But this whole thing about accounting for multiple
flaws -- remember, each vessel may have three, four, five
thousand flaws, and you go through that loop and you get
values of conditional probability of initiation for flaw
number one, flaw number two. Actually, the way it seems to
turn out, maybe out of that 3,500, maybe four or five of
them will be non-zero, okay?
So, the question is now, for that vessel, what's
the probability of initiation, and I'm not going to go
through all this equation-looking stuff, other than to say,
if CPI is the conditional probability of initiation, one
minus CPI is the probability of non-initiation, and then, if
you -- for two flaws, if you take one minus CPI for the
first flaw and multiply it times one minus CPI for the
second flaw, what you have is the probability that neither
one of those flaws initiated, you have a joint probability
that neither one of those flaws initiated, right, and if you
have 3,000 flaws, it's still just one minus the CPI times
one minus the CPI all the way out to however many flaws you
had.
So, at the end of that, that's the probability
that none of those flaws initiated. Then, if you subtract
that from one, it's the probability that at least one of
them did initiate it.
That's the value -- that's what this is an attempt
to show. That's the value that goes into your PFM-I array
for that vessel transient, that IJ entry in your PFM-I
array.
So, you go through that business about how did K1
penetrate K1C space, you get a value of CPI for that flaw,
you do it for many, many flaws.
Then you do this and you get a value to go into
your PFM-I matrix.
One other little -- this max in here -- you do it
for the maximum value as a function of time for each flaw.
In other words, take the peak value.
So, for this particular flaw, you know, let's say
this was the case.
We would come out here and we would take that
value for flaw number one, and then if, you know, we had
another non-zero, we would do the one minus that time one
minus that to get the value that goes into the PFM-I array.
DR. KRESS: Multiply that probability times the
time?
MR. DIXON: Not times the time. That's the
conditional -- each one of these is instantaneous, but if
you think about it -- that's a good question. This value
here really is the cumulative value of everything that's
gone before.
DR. KRESS: I'm trying to get a probability
density function integrated over time, but I don't see how
to do it.
MR. DIXON: That's not what's going on here. I
know it's a lot to get your fingers around at one time.
I'll just conclude with, you know, one that I
showed earlier. You know, the goal is to have the code
ready to go by March 1, 2000. This assumes, you know, that
all the models are finalized according to schedule.
In the interim period, we're going to finalize
some models, implement models in the FAVOR, and perform
scoping studies, and it looks like Oconee will be the unit
that's the guinea pig for the scoping studies, because the
thermal hydraulics and the PRA are going to be finished.
That's it. That concludes everything that I have.
SPEAKER: [Inaudible.] Pieces of this don't break
off all that easily.
DR. KRESS: No, it doesn't seem to.
SPEAKER: What do you think is important for the
rest of the committee to hear out of this, to let them know
where the staff is, possibly raise questions about the
recommendations on where they should go?
SPEAKER: George seemed very concerned about the
uncertainty analysis in the K1A.
DR. KRESS: Terry walking through that thing would
bring that out, I think, would be one of the things.
SPEAKER: That might be my candidate.
DR. KRESS: Yeah. I was about to say that would
be my candidate.
SPEAKER: And hold off on the flaw distribution
until they're ready with a final report, although I would
have thought it was going to go the other way around.
DR. SEALE: I think something about how they plan
to integrate the PRA data into FAVOR -- that is, the PRA
process -- what they expect to have as a communication
vehicle in order to get the risk-based output.
DR. KRESS: Terry had one slide on that which
would cover it, I think.
SPEAKER: I really think these two pieces are the
ones that maybe --
DR. KRESS: Which is that?
SPEAKER: This fracture toughness uncertainty with
the RTNDT.
DR. KRESS: Yeah.
SPEAKER: Because it sort of puts those pieces
together.
SPEAKER: That's with an understanding that there
will be a more detailed, updated meeting on the
uncertainties?
SPEAKER: Certainly, the whole uncertainties, but
at least to give us the chance to go through the mechanics
of what we're doing.
DR. KRESS: I'm quite interested in this risk
acceptance criteria, 1 times 10 to the minus 6, but I can't
see that there's anything they can present to us at the next
meeting for that. I mean somebody is working on that and
thinking about it. We didn't hear anything today about it.
SPEAKER: I guess I'd agree with Dr. Kress. I
don't think we're ready to talk about it. My understanding
is there's some work going on there, but we won't be ready
for that.
I guess I'd agree with Bill on those two pieces,
with one caveat, I guess.
I know Nolan, Nathan, and I were talking
separately that to do the meeting that I think Professor
Apostolakis was asking for, we may not be quite ready for
that till maybe December timeframe, to really spend a day on
uncertainty and track through that, but I think we could do
a reprise of those -- you know, Terry's and Mark's
presentations, maybe trying to articulate some --
SPEAKER: I sort of realized you weren't going to
be ready to do the full uncertainty. It was just a question
of what we could do sort of leading up to that and, I think,
highlighting some places where it seemed especially
uncertain how to handle the uncertainty.
SPEAKER: Yeah, and I guess our reluctance, a bit,
is because this is work in progress.
SPEAKER: That is the problem here, that
everything is work in progress.
SPEAKER: Right.
DR. SEALE: Not that we don't like to be able to
put our finger in the soup while it's still fresh.
SPEAKER: No doubt.
SPEAKER: You know, I am a little concerned, you
know, with Tom's question that, you know, we're raising a
fairly fundamental issue about the acceptance criteria, you
know, can we work from the LERF goal in 1.174.
Do we need to formally somehow get that raised for
staff consideration, or do we consider it raised at this
point?
SPEAKER: It's certainly been raised. I mean I
believe the SECY paper recognized that this was an issue
that had to be dealt with.
We said that we were going to do a scoping study
that would --
SPEAKER: But the SECY paper really started with
the 1.174 criteria as the ultimate acceptance criteria.
SPEAKER: That may be. I was under the
impression, speaking with Mark, that there were still
questions about that. That was certainly a model of how
we're going to proceed. It was not necessarily the only
model that we were going to look at.
I thought that that was part of the discussion on
-- once we apply -- we tried to apply some of the latest
thoughts on how we're doing the risk-informed applications,
whether or not we'd come back to PTS and say, okay, now we
need to look at things a little bit differently now. I
thought that was open under the SECY.
Anyway, if the SECY didn't say that, we're not
saying that's necessarily the ultimate goal.
DR. SEALE: Certainly, the one size fits all is
not the right way to go because of this question of the
containment and issues like the spent fuel fire and so on.
DR. KRESS: It's the same issue.
DR. SEALE: It's the same issue, but it shows up
in very specific examples.
SPEAKER: Understood.
SPEAKER: That may be a judgement call, but that
may be worth some more discussion. We had a meeting for the
RES division directors, for Farouk and Tom King and Mike
Mayfield, where we talked about, you know, fleshing out this
issue of the containment integrity in LERF.
Obviously, the committee has weighed in on that
already once and, I think, weighed in on the side of we'd
like to see the staff take that on, is what I recall.
SPEAKER: I'm not sure Tom's issue came up in that
discussion.
DR. KRESS: I doubt if it came up then.
SPEAKER: What we don't want to do is raise this
six months from now. I just want to make sure that it gets
-- you know, the notion that, you know, the source term that
was used to generate that LERF may not be the right source
term for the PTS.
DR. SEALE: We may need to highlight it.
SPEAKER: Widely different situations.
SPEAKER: Does that address your concerns that we,
you know, somehow have to get that into a letter or a formal
presentation of the committee?
SPEAKER: I think the committee needs to think
about what message and what way they're going to transmit
it.
SPEAKER: And we haven't really raised this issue
with the full committee either.
DR. KRESS: We can't do it as a subcommittee. It
has to be the full committee. That may be a subject we
might want on the full committee agenda, even though they're
not ready to talk about it.
SPEAKER: You can just have a few minutes to raise
that concern.
DR. KRESS: Okay. Let's do it that way. I'll
raise the concern.
SPEAKER: The staff is not ready to address it,
but you know, it's a concern that we've raised.
DR. KRESS: That way we'll raise it to the level.
SPEAKER: So, you're not looking for a staff
presentation on that.
SPEAKER: No.
SPEAKER: Unless you're ready.
SPEAKER: At least philosophically, just to go
around sort of what the division directors were talking
about the other day, I believe Farouk or Dave Bessette have
talked about they've tasked Professor Diafanis with looking
at containment pressurization and any failures that may
result from containment pressurization due to PRS, and then
Mike Mayfield chimed in with the thing we talked about at
the beginning here, that I'm not overly worried about
containment pressurization, I'm worried about this
displacement of the vessel.
SPEAKER: But see, that all relates to containment
failure, and Tom's concern is, once the containment fails,
you know, what's an acceptable probability that you have a
different consequence.
SPEAKER: And again, I think a number of folks
have raised different issues, and different people on the
staff have different opinions as to what's going to happen
or how this will be addressed, and we're clearly not ready
to talk about that in any consistent way.
SPEAKER: If the committee has some
recommendations on how to proceed, I think it would be
worthwhile hearing.
DR. KRESS: Well, I can maybe suggest something,
but what I'll plan on doing is articulating the concern to
the full committee, and that will raise it.
SPEAKER: So, we'll have the presentations, then,
on the -- the two presentations.
SPEAKER: What do we have, two hours, Noel?
SPEAKER: Two hours. We don't need to use the
whole amount.
SPEAKER: Okay. We'll try and come in with
shortened versions of these.
DR. SEALE: Dana will figure out what to do with
anything you give the committee.
SPEAKER: Somehow, I suspect, with Professor
Apostalokis, I wouldn't count on shortening it too much.
DR. KRESS: I'd shorten it, but I wouldn't count
on shortening it two hours. I'd shorten the presentation.
SPEAKER: I'd shorten the presentation, but I
wouldn't take too much of the time back.
DR. KRESS: That's right.
SPEAKER: Okay. Sounds good.
SPEAKER: Thank you.
DR. KRESS: As usual, a very professional
presentation. We appreciate it.
SPEAKER: Could we get a copy of the most recent
version of the generalized flaw distribution paper, since
the one we have seems to be a somewhat out of date version?
SPEAKER: Yeah, I guess I should have summarized,
because I knew George had asked for the P.D. Ruff NUREGs,
basically, volumes one and two are available publicly now,
and also the reports on Prodigal. So, Debbie took an action
to get those, and I guess we can get them.
SPEAKER: I have those, but I don't know whether I
get those as ACRS or UC-5. You never know how they're
coming in.
[Inaudible conversation.]
END OF TAPE 4, SIDE B
Page Last Reviewed/Updated Tuesday, July 12, 2016