Reliability and Probabilistic Risk Assessment - April 17, 2001
Official Transcript of Proceedings
NUCLEAR REGULATORY COMMISSION
Title: Advisory Committee on Reactor Safeguards
Reliability and Probabilistic Risk
Docket Number: (not applicable)
Location: Rockville, Maryland
Date: Tuesday, April 17, 2001
Work Order No.: NRC-157 Pages 1-205
NEAL R. GROSS AND CO., INC.
Court Reporters and Transcribers
1323 Rhode Island Avenue, N.W.
Washington, D.C. 20005
(202) 234-4433. UNITED STATES OF AMERICA
NUCLEAR REGULATORY COMMISSION
+ + + + +
ADVISORY COMMITTEE ON REACTOR SAFEGUARDS
MEETING OF THE
SUBCOMMITTEE ON RELIABILITY AND PROBABILISTIC
+ + + + +
April 17, 2001
+ + + + +
+ + + + +
The Subcommittee met at the Nuclear
Regulatory Commission, Two White Flint North, Room T-
2B3, 11545 Rockville Pike, at 8:30 a.m., Doctor George
E. Apostolakis, Chairman, presiding.
GEORGE E. APOSTOLAKIS Chairman
MARIO V. BONACA Member
THOMAS S. KRESS Member
GRAHAM M. LEITCH Member
ROBERT E. UHRIG Member
ACRS STAFF PRESENT:
MICHAEL T. MARKLEY
TOM BOYCE NRR
STEVE EIDE INEEL
ADEL EL-BASSIONI NRR
TOM HOUGHTON NEI
ROGER HUSTON Licensing Support Services
MICHAEL R. JOHNSON NRR
STEVEN E. MAYS NRR
DEANN RALEIGH LIS, Scientech
JENNY WEIL McGraw-Hill
TOM WOLE RES
BOB YOUNGBLOOD ISL
Agenda Item Page No.
Review goals and objectives for this meeting;
past ACRS deliberations on risk-based
performance indicators (RBPIs), G. APOSTOLAKIS . . 5
NRC Staff Presentation
Background/Introduction, S. MAYS, RES. . . . . . . 6
Relations of RBPIs to Revised Reactor
Oversight Process (RROP), TOM BOYCE, NRR . . . . . 9
Potential benefits. S. MAYS, RES . . . . . .50
RBPI development process, S. MAYS, RES
H. HAMZEHEE, RES
Summary of results, S. MAYS, RES . . . . . . . . .63
Initiating Events: full-power/internal,
H. HAMZEHEE, RES
Mitigating systems: full power/internal. . .98
Containment. . . . . . . . . . . . . . . . 108
Shutdown . . . . . . . . . . . . . . . . . 120
Fire events. . . . . . . . . . . . . . . . 132
Industry-wide trending . . . . . . . . . . 134
Risk coverage. . . . . . . . . . . . . . . 136
Verification and validation results. . . . 143
AGENDA - (Continued):
Agenda Item Page No.
NRC Staff Presentation - continued
Discussion of implementation issues, . . . . . . 150
S. MAYS, RES
H. HAMZEHEE, RES
Discussion of industry comments
S. MAYS, RES
H. HAMZEHEE, RES
Industry perspectives on RBPIs,
T. HOUGHTON, NEI . . . . . . . . . . . . . 178
General Discussion and Adjournment
General discussion and comments by members . . . 191
of the Subcommittee; items for May 10-12,
2000 ACRS meeting, G. APOSTOLAKIS, ACRS
CHAIRMAN APOSTOLAKIS: The meeting will now
come to order. This is a meeting of the Advisory
Committee on Reactor Safeguards Subcommittee on
Reliability and Probabilistic Risk Assessment. I am
George Apostolakis, Chairman of the Subcommittee.
Subcommittee Members in attendance are Tom
Kress, Graham Leitch, and Robert Uhrig, and Mario
The purpose of this meeting is to discuss
the results of the staff's Phase 1 effort to develop
risk-based performance indicators. The Subcommittee
will gather information, analyze relevant issues and
facts, and formulate proposed positions and actions,
as appropriate, for deliberation by the full
Committee. Michael T. Markley 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 March 26, 2001.
A transcript of the meeting is being kept
and will be made available as stated in the Federal
Register Notice. It is requested that speakers first
identify themselves and speak with sufficient clarity
and volume so that they can be readily heard.
We have received no written comments or
requests for time to make oral statements from members
of the public regarding today's meeting.
We will now proceed with the meeting and
I call upon Mr. Steve Mays to begin.
MR. MAYS: Thank you, George. I'm Steve
Mays from the Office of Nuclear Regulatory Research.
With me today at the front is Hossein Hamzehee, who is
the Project Manager for working on risk-based
performance indicators, and with me also to my left is
Tom Boyce from the Office of Nuclear Reactor
Regulation, who will speak in a couple minutes about
the relationship that this work has to the Reactor
Oversight Process. Also here at the side table is
Mike Johnson, who is the Section Chief in NRR, who is
our technical counterpart in NRR and our liaison with
this work, and we have a couple of our contractors in
the audience if there's any questions that I can't
directly answer or Hossein can't answer, they can come
up and give additional information about what we've
What we are trying to do today is give the
ACRS an opportunity to provide some comments and to
provide some information to you about what was in the
report that we issued in January. We've already held
one public meeting in February to kind of lay out
what's in the report, and kind of frame the discussion
of what we are trying to do, and how we went about
doing it, so that when we have our public meeting next
week we would have the opportunity to make sure that
was a well-focused meeting and directed towards the
kinds of things we need to know what the response from
the outside stakeholders is.
We extended our comment period at the
request of people in the February meeting to May, so
that people can come to the meeting next week, discuss
points, get the opportunity to hear some answers from
us if they do, and then be able to take that into
consideration as they give us their formal comments in
May. We are looking forward to that meeting, and part
of what we are going to see here today is some new
stuff that we've done that's not actually in the
report, and we are going to also present that at the
meeting next week, so that we can hopefully move this
So, we are looking for feedback from the
ACRS, we expect probably a letter of some kind with
respect to whether they believe that the work we are
doing is a potential benefit to the Reactor Oversight
Process, whether we've gone about that in a
technically sound manner, and also to get some
feedback on the alternate approaches that we're going
to present today, which are not in the report, that
we've gone off and developed in light of some of the
early comments we got, both internally from the NRC
review, as well as some comments we've had from
external stakeholders relating to the total number of
potential indicators and what that impact would be on
the oversight process.
So, we are going to have this briefing
broken up into several pieces. The first part is the
relationship of the RBPIs to the Reactor Oversight
Program. Tom Boyce from NRR to my left will be
discussing that. Then I will come back and the rest
of the presentation will be primarily from our part on
the technical aspects of what's been done, including
what we see as the potential benefits, what we
actually did in development, some results that we
have. We want to go over the key implementation
issues that are before us, because we think those tend
to be the ones that we have the biggest comments on
from both internal and external reviewers so far, and
to go over the alternate approaches that we're looking
at as a means of dealing with some of the issues that
have been raised.
So, with that, I would like to go to Tom
Boyce, who will discuss the relationships of the RBPIs
to the Reactor Oversight Process.
MR. BOYCE: Good morning.
As Steve said, I'm Tom Boyce, I'm in the
Inspection Program Branch in the Office of NRR, and
NRR requested that we have a short amount of time at
the beginning of Research's presentation to let you
know the relationship, as NRR sees it, of the risk-
based PI development program to the current
performance indicators in the Reactor Oversight
CHAIRMAN APOSTOLAKIS: Is your presentation
consistent with the memorandum from Mr. Dean to Mr.
King, of December 1, 2000?
MR. BOYCE: It is entirely consistent. I
was the author of that memo.
CHAIRMAN APOSTOLAKIS: Okay, good.
MR. BOYCE: By definition.
CHAIRMAN APOSTOLAKIS: You may have changed
MR. BOYCE: I maybe shouldn't have stated
it quite so positively.
Before I talk about the Reactor Oversight
Process relationship to risk-based PIs, it's important
to understand the overall environment with which our
agency is now regulated, and some of the changes that
are impacting the nuclear industry.
The Commission has provided direction to
the staff that its intent is to better risk inform the
NRC's processes, and it's done this for several years
on a variety of fronts. The Reactor Oversight Process
was revised in 1999 to be more risk informed,
objective, understandable and more predictable than
the previous oversight process. The Reactor Oversight
Process was implemented on April 2, 2000, so we have
had one year of practice in the Reactor Oversight
Process under our belts.
Another backdrop for the industry is
continuing advances in the use of information
technology and data. Industry is getting better and
better at collecting data, processing it for its own
internal uses. We also are getting better at it. The
Reactor Oversight Process has got a web site that has
gathered a great deal of kudos for its ability to
The internet and PCS have allowed much
more free exchange of information than has previously
been allowed, and both NRC and industry are continuing
to expand their capabilities in this area.
We wrote about the bases for the Reactor
Oversight Process in two Commission papers in early
1999, and there we stated that the Reactor Oversight
Process would use a combination of inspection findings
and performance indicators to provide oversight of
industry. We conducted a pilot program in 1999, and
the results were articulated in SECY-00-049. In that
same Commission paper, we stated that while the future
success of the Reactor Oversight Process would not be
predicated on the risk-based PI program, we thought
that there were a couple of places where the risk-
based PI program could, in fact, enhance our current
set of performance indicators.
These areas are actually articulated in
the last bullet right here, the reliability
indicators, unavailability indicators, shutdown and
fire indicators, and containment indicators.
We also thought that the risk-based PI
program offered the potential to establish, perhaps,
plant specific thresholds for these PIs on the current
set of PIs. Because we thought this, we decided to
task research to develop in these areas, and we sent
them a user need. Research responded that they would
examine the feasibility of these selected risk-based
PIs as part of their Phase 1 report, and you'll be
hearing more about that in a little bit.
Even though the risk-based PI program is
moving forward, we thought that there were several key
implementation issues that needed to be addressed
prior to implementing the risk-based PIs
incorporating them into the Reactor Oversight Process.
One of the keys, in general, is data quality and
availability. Our experience in the Reactor Oversight
Process is, is that while data is being collected by
individual licensees there are a variety of ways that
you can collect that data. There is a variety of
quality for that data, and how you collect that data
and pull it together into a graph that is presentable,
we found surprising variation. So, we thought that we
needed to be happy with the way data was collected, so
that it was done uniformly and consistently, before we
are able to implement it in the Reactor Oversight
Second, we thought that the models used
for assessing the data needed to be developed and
validated by licensees and the NRC staff in the
regions, and you'll hear more about the status of
development of SPAR models from Steve, but those two
were the key areas that we thought needed to be fully
mature before it was ready to be incorporated in the
Reactor Oversight Process.
CHAIRMAN APOSTOLAKIS: I'm a bit confused
now. Isn't this, aren't these two applicable to the
existing revised Reactor Oversight Process? I mean,
you also need good data, you also need some sort of a
PRA, to assess the significance of a particular
performance indicator being above a number and so on.
So, I don't know, why are these two issues,
implementation issues, so important to risk-based
performance indicators, but not to the existing
MR. BOYCE: Well, in the case of the
existing performance indicators for the ROP, we had an
opportunity to go through a pilot program, licensees
submit the data directly to the NRC, using mutually
agreed upon guidelines in the NEI document, NEI 99
Tech 02. That was developed by mutual discussions
with industry, over an extended and intensive an
extended period of time in an intensive manner. The
current data for the risk-based PI program is drawn
from sources such as the EPIX database, that's, I
believe, under the auspices of INPO, and that same
sort of rigorous look at how the data is submitted has
not been applied yet.
CHAIRMAN APOSTOLAKIS: But, it could.
MR. BOYCE: It could.
CHAIRMAN APOSTOLAKIS: It could, I mean,
they could do the same first of all, I don't like
this idea of they and us, I mean, it's one agency, but
there is nothing in the methodology that says, you
know, you have to use EPIX.
MR. BOYCE: Correct.
CHAIRMAN APOSTOLAKIS: They can use the
data that you are using.
Now, they felt the need to go to other
sources of data, because for some reason the data that
we receive right now is not sufficient, is that the
MR. MAYS: Yes, George. Let me propose
we'll get into that in much more detail in the section
where we talk about implementation issues, but to
address it shortly, remember when the ROP was put in
place one of the key issues for getting indicators was
what information is readily available and can be put
together in a consistent way, and that data that's
reported into the ROP is reported under 50.9
requirements for licensee submittal of data. That's
one of the major issues with respect to implementation
of these PIs, and that data that's being submitted
under the ROP was not specifically tailored to certain
aspects, like reliability indicators, and the models
in the ROP were more generic with respect to the
So, there were several things of that
nature that I would put in the category of expediency,
that required that to be there, and as we move to more
detailed and more plant-specific data and thresholds,
we think it's important to make sure that we have an
understanding of what that data is and a common
agreement as to how the quality of the data needs to
be assured and how that stuff needs to be reported,
and that's an implementation issue we're going to have
to work out, but generically, as long as you have the
data that fulfills the model, then that's all you
really need from a modeling standpoint and a
calculational standpoint, but from a regulatory
standpoint there are other issues that have to be
MR. JOHNSON: And, if I could add, Michael
Johnson, NRR, if I could add to what Steve has said,
and I think he's made some good points, remember the
challenges that we face with the ROP PIs, as we'll
talk about when we brief the ACRS in May, during the
first year of implementation have been challenges
associated with verification of the data, even with
the relatively simple PIs that we have now in the ROP,
it's a problem.
So, George, to go to your question, your
point, it's not that we don't face these challenges,
these similar challenges with the existing ROP, it's
that these challenges will certainly exist as we go
forward with RBPIs.
CHAIRMAN APOSTOLAKIS: Yes, they do exist.
Well, I read the memorandum that I
mentioned earlier, dated December 1st, from Mr. Dean
to Mr. King, and I must admit I was surprised at how
cool it is towards this effort, as if somebody is
trying to force this upon you and you are resisting.
The report did not demonstrate that the
proposed RBPIs will be more effective than the PIs
currently in place. I don't know what that means.
Licensees may be reluctant to voluntarily implement
the new RBPIs because of two reasons, there are many
more indicators to track, calculate and report, which
increases the effort licensees have to expend. So
what, if they have to do it, they have to do it.
Where is the technical argument? Is there any
justification for needing more indicators to track,
calculate and report? That should be our criterion,
that there is some information there that's useful to
us, not that it imposes burden on the licensees.
First, we have to decide whether it's unnecessary, and
if it's unnecessary then, of course, we don't impose
it. But, I didn't see any argument anywhere here that
says, no, these additional indicators are not needed
because we already cover them. It just says, you know,
the licensees will have to spend more time doing it,
and, boy, we really don't want to do that, and
licensees will be putting themselves in a position
where it is much more likely they will have to report
a non-green PI and subject themselves to the resulting
increased regulatory and public attention. Well, I'm
shocked, I'm shocked shocked. There are indicators
that are not green? I just don't understand this
You guys don't like something, but you
don't want to come out and say it to us. Obviously,
you don't like it.
MR. JOHNSON: Steve, when we get into
under the implementation issues, will we come back to
MR. MAYS: I think we will.
In fairness to in fairness to Tom and
building shock, we'll put that in.
CHAIRMAN APOSTOLAKIS: I just didn't want
to put Tom on the spot, but he's the one here.
MR. MAYS: I know.
MR. BOYCE: Thank you, George. I don't
CHAIRMAN APOSTOLAKIS: And, he said he
wrote it, big mistake.
MR. BOYCE: I retract my earlier comments.
CHAIRMAN APOSTOLAKIS: Okay.
MR. MAYS: Tom hasn't been before the ACRS
as often as I and Mike have.
In fairness, I think we, in the RBPI
report, raise the issues of the implementation because
we recognized that if we were going to make an
improvement in the ROP it was going to be a voluntary
improvement that we decided we wanted and that we
negotiated with our external stakeholders to determine
it was a benefit to the agency, and I think what you
were seeing there was just a recognition of some of
the issues that we knew were going to be raised, and
that we knew had to be addressed, as opposed to saying
that they could not, or should not, be done here. I
read that letter when I saw it more as a confirmation
that we had identified the correct issue in the RBPI
document, and that we knew from our previous
interactions with external stakeholders that those
were going to be concerns that we had to address, and
that the Commission was the one who was eventually
going to adjudicate whether or not we were doing that
properly or not.
So, I don't think it was nearly as
negative as you might have portrayed it.
CHAIRMAN APOSTOLAKIS: I thought it was
cool, there is a certain coolness here, that maybe
what you guys are doing has some value, but you have
not demonstrated it to us, and what's worse, you may
ask the licensees to do more.
MR. JOHNSON: Well, there is an aspect of
that, and maybe Tom was going to get into that. Let
me just say a couple of words before Tom.
CHAIRMAN APOSTOLAKIS: You should let him
at some point defend it.
MR. JOHNSON: Yes, we should.
MR. BOYCE: No, go ahead, Mike.
MR. JOHNSON: As Steve sort of indicated,
there is an aspect of this, and the ROP, when we set
out to develop performance indicators, remember the
performance indicator aspect of the ROP is a voluntary
program, if you will. Even the document that endorses
the guidance is the guidance is an NEI document,
NEI 99-002, that provides the criteria, we endorse
CHAIRMAN APOSTOLAKIS: How many licensees
MR. JOHNSON: None of the licensees have.
CHAIRMAN APOSTOLAKIS: It is still
MR. JOHNSON: Yeah. All of the licensees
are reporting on their existing PIs.
And so, what we are talking about with
risk-based performance indicators, as Steve indicated,
is an enhancement to this PI reporting program that's
a piece of the ROP, and as such, I mean, I think we
do, in fact, need to be careful about things like, are
we increasing the burden without commensurate benefit?
In fact, in our formal change process for the
performance indicators, we look at, should we be
making reductions in the inspection program, or
changes in the inspection program, in areas where we
have information that we get readily from the
performance indicators, so all of those things have to
be worked out in the implementation stages. So, we
wouldn't just adopt a suit of PIs that would make us
happy, if you will, without regard to the impact that
they would have on licensees.
CHAIRMAN APOSTOLAKIS: And, I think that's
a very reasonable thing to do, as long as there is
also a technical discussion
MR. JOHNSON: That's right.
CHAIRMAN APOSTOLAKIS: as to, you know,
this indicator gives us information we already have,
or maybe expands on something, but by and large we
really understand what's going on and the additional
burden is not justified.
MR. JOHNSON: That's right.
CHAIRMAN APOSTOLAKIS: I can see arguments
like that, but just to say that, you know, this
imposes burden, without addressing the kind of
information you get, I find that a little odd.
MR. BOYCE: I think there was also a
meeting that followed that memo, that was held
between, I think, Sam Collins and members of the
Research staff, and I think there we were able to get
past some of the detailed discussion you saw at that
memo, and I think at that meeting we said that we
believed that this was a good technical effort, it did
have potential value, and we did want them to continue
development, which is not stated explicitly in that
memo, because the intent of that memo, as I recall,
was to convey technical comments on the report itself.
CHAIRMAN APOSTOLAKIS: Now, I'm just
curious, what would you expect them to do to
demonstrate that the proposed RBPIs will be more
effective than the PIs currently in place? The word
"effective," what does it mean in this context, I
mean, independently of the coolness of this, I mean,
technically, what would you expect them to do?
MR. BOYCE: Well, I'm not sure we've
established hard criteria for what we mean by more
effective, but, in general, the PIs that we have have
certain limitations. I mean, not all of them have
been well-founded and risk-informed principles. Some
were selected based on 95 percent performance of
industry, there's a word, but it's not
probabilistically-based, it was, we took a look at
histograms and said that 95 percent of the plants
operate below this threshold.
CHAIRMAN APOSTOLAKIS: So, this is
MR. BOYCE: Thresholds.
CHAIRMAN APOSTOLAKIS: Yes.
MR. BOYCE: So, we could certainly improve
on our technical basis for thresholds for individual
PIs, that's one area.
CHAIRMAN APOSTOLAKIS: And, there is a nice
criticism of that on page A-10 of Appendix A, very
nice. It says, "If I wrote ..."
MR. BOYCE: But, we couldn't do that,
George, that would be a conflict of interest.
MR. JOHNSON: Can I add one other thing to
that answer, is that, again, I'll allude to a change
of the formal change process. What happens at the end
of this effort, and what happens, in fact, when we go
to put in place any new PI, as we go through a formal
change process, and that process has astute things,
like we'll conduct a pilot, we set criteria up at the
beginning of that pilot for what we want to see in
terms of evaluating the efficacy of these proposed new
PIs, and so, it's in those criteria that we'll be very
specific about what we'll look for in terms of making
a decision about whether to go ahead.
And, there's something already we'll
talk to you again in May about two PIs that we already
are piloting under the existing ROP that are not risk-
based PIs, but they are going through the process. We
have a pilot in place, we are looking at the results
of that pilot. We are looking at the performance as
would be indicated by indicators reported against
those proposed PIs, balancing that against the
existing PIs to see if there are differences. So,
it's those kinds of things that you look at, and those
are built into this formal process that we enter into,
after this phase of this preliminary development of
the RBPIs is finished.
MR. BOYCE: And finally, one more comment
on the tone of that memo. We had gotten informal
feedback from some stakeholders that the risk-based PI
program had, through whatever means, been perceived as
a certainty, that it would, in fact, be implemented.
And, we wanted to make sure that that expectation
was, in fact, addressed so that it would be put in the
right context. In other words, the change process
that Mike just alluded to did need to be followed.
There are, I think, 30 some odd performance indicators
that are being proposed here in the Phase 1 report,
and the data collection requirements do, in fact, add
significant burden to licensees. So, licensees do
give us a feedback that, hey, if you are going to
implement the new program like that, you need to
consider cost benefit, and we had not even engaged in
terms of cost benefit at that point. So, to some
extent, the tone you see there was to try and address
the perception that had gone out in industry.
DOCTOR KRESS: If you implement this, would
you do it on a pilot basis with a number of volunteer
plants to start with?
MR. BOYCE: We would expect that to be the
case, that's what our Manual Chapter 0608 calls for
for PIs like this.
DOCTOR KRESS: And, to determine whether or
not it's useful, then those volunteer plants would
have had to been compared with the old program, and
would have had to have degraded performance somewhere,
otherwise you are proving a negative.
I'm not sure, you know, you might go on
for years, and years, and years, before you ever come
to some conclusion that the new process is useful to
you that way.
MR. BOYCE: Well, I think you'll see from
the report that Steve is going to go over that that's
not, in fact, what happened. I think they ran some
test data through and found out that there was, in
fact, degraded performance that came out for the set
of data that they looked at.
DOCTOR KRESS: Oh, you looked at highest
MR. BOYCE: Well, let me we are jumping
the gun a little bit.
DOCTOR KRESS: Okay.
MR. BOYCE: We looked at the performance
over the '99 time period, basically, '97 through '99
time period, which is a time period for which the ROP
pilot program and the ROP program already had some
data on plants. So, we have looked at that, and we do
know that because there are certain areas that we are
examining with PIs that the ROP doesn't have PIs that
we now have the opportunity to see things that weren't
there potentially as indicators in the ROP.
We are going to cover some of this stuff
when we talk about potential benefits and things of
that nature, and the examples you'll see when we ran
for the 23 plants that we did run, we do have a fairly
broad range of coverage of that.
DOCTOR KRESS: Okay.
Could you indicate what are the
implications when you say reporting under 50.9, I'm
not sure I'm exactly familiar, I have kind of an idea,
but exactly what does that mean?
MR. BOYCE: 50.9 requires that information
submitted to the NRC by the licensees will be
submitted under oath and affirmation, and that means
when a utility does that, that information, once
submitted, can be cited in violations for failure to
be accurate. So, when you have that level of rigor
applied to data, and the potential for being cited for
inaccuracies in that data, that process makes
utilities apply more effort to ensure that that same
data is correct than they may otherwise have to, and
that's one of the data issues we'll get to.
DOCTOR KRESS: Okay.
MR. BOYCE: And, I'll give you some
examples of things when we get to that area, as to how
that can be a potential problem.
And, the issue from our standpoint is,
what level of quality and rigor does the data have to
have, and if that quality and rigor is something
different from 50.9 then the question is, is, well,
why would we have to, or would we have to have data
submitted under 50.9. That's an implementation issue
that would have to get addressed through the formal
process in the ROP change process that Mike and Tom
just alluded to.
In fact, when the existing ROPs were first
being tried, one of the things that happened, in order
to make sure they understood what the quality issues
were and the difficulties were, was there was, I guess
the right word would be, a waiver
DOCTOR KRESS: A discretion.
MR. BOYCE: a discretion on enforcement
on those issues, as part of the initial program, to
make sure that that wasn't becoming an impediment to
testing the program out and understanding what levels
of things needed to be done.
So, those are all kinds of details of how
you would go about doing the implementation, which,
quite frankly, we're not really here to discuss
exactly how that will happen today. The process would
be more like this. We go through public comment, we
get ACRS comments on the technical quality of what
this program brings, and whether or not it looks like
it's beneficial to the ROP, and at that point we would
produce our Phase 1 report and NRR at that point would
be in a position of saying, do you want to take all of
these, some of these, none of these, and try to run
them through the ROP change process. And then, once
they would go into that process, they would lay out
the plans in accordance with that procedure, and get
together with the industry and our other external
stakeholders, and go through the process.
So, it's a little premature to tell you
what that all would have in it, or what all the
decisions would be made, and how they would all be
made, because that's a little bit ahead of where we
want to be right now. We want to
DOCTOR KRESS: I'm just trying to
understand industry's concern. I guess by extension
then, all the EPIX data then could theoretically be
subject to the requirements of 50.9.
MR. BOYCE: Could be. It's not certain
that they would, it's not certain that they would not,
that's an implementation issue we have to address, and
we have recognized that that was a significant issue
when they were doing the unavailability data for the
current ROP indicators, and we no reason why that
issue wouldn't also be an issue for reliability data,
which is what the EPIX data is being used for here.
So, we recognize that that's an issue that
has to get resolved. Industry recognizes it as an
issue, and we all think it's something that has to be
taken care of through the ROP change process.
CHAIRMAN APOSTOLAKIS: So, the current
oversight, revised oversight process, when it does
risk-related calculations what models is it using?
MR. BOYCE: Well, actually, it's not doing
risk calculations in the PIs. The calculations were
done initially to establish what the thresholds should
be on the PIs.
CHAIRMAN APOSTOLAKIS: And, that's it.
MR. BOYCE: Now, after that, all they do is
calculate the value that's coming in and compare it to
the threshold. There's no more risk modeling being
done in the ROP to get the current indicators.
CHAIRMAN APOSTOLAKIS: But, you are doing
the same, aren't you?
MR. BOYCE: We are applying the same
CHAIRMAN APOSTOLAKIS: Okay, but you are
using the SPAR model.
MR. BOYCE: We are using plant-specific
SPAR models to set thresholds.
CHAIRMAN APOSTOLAKIS: What did they use?
MR. BOYCE: They used a combination of
licensee models, some SPAR model runs that we did for
them in the process, and they came to a consensus
opinion of how to set thresholds based on those
results. And, they tend to be generic for the
industry, as opposed to plant specific.
And so, that's what was done, and it's
documented in 99-007. I can't recall exactly which
appendix it's in, but I know it's in there.
CHAIRMAN APOSTOLAKIS: It was H, Appendix
I'm trying to see whether I mean, the
sig I understand the significance of the first sub-
bullet, data quality and availability, the second one
is not so clear to me. Is it because this new effort
is intended to be plant specific?
MR. BOYCE: That's right.
CHAIRMAN APOSTOLAKIS: The PRA model is
more important than it was in the generic case.
MR. BOYCE: That's right.
You see, the thresholds for individual
plants would could be lowered, and so the plant's
margin to the green/white threshold, if we use the
same process under the current ROP, could be less.
And, any time you are talking about increased
regulatory attention licensees are very sensitive to
that sort of thing, and so we want to have good
quality models so we have confidence in the thresholds
and in the information that's being presented to us.
CHAIRMAN APOSTOLAKIS: Right.
And, my counter argument to that is that,
this is a good idea to worry about these things if the
starting point has some logic to it. And, I don't
think that 95th percentile you guys did can withstand
MR. MAYS: Well, George, you'll notice that
one of the things
CHAIRMAN APOSTOLAKIS: The weakness of this
method is that it depends only on the number of plants
with less than acceptable performance, but not on how
much their performance exceeds the norm. Wonderful.
MR. MAYS: Well, George, we went back, as
part of the RBPI process, as we outlined in the RBPI
White Paper, and said we were going to look at how we
thought thresholds need to be set, and we went and
looked at that particular issue and we, in
recommendations in the program, concluded that we
thought it made more sense to have the green/white
threshold for performance indicators be based on a
risk change rather than on a deviation from norm
principle. And so, we have made that case that it's
more consistent with the significance determination
process for inspection findings, which all three color
layers are based on a risk metric, and so we are
making that recommendation, we provided the
information for how we would say what the distribution
of plants' performance was, where the 95th percentile
was on that, and we've made that recommendation. And,
so far, quite frankly, I haven't had any technical
comments come back from either inside NRC or outside
so far, that said, no, no, no, we want to stay with
the 95th. There may be some that will say that, but
I think there is a better logical connection for using
the green/white interface on PIs based on risk than
based on deviation from the norm.
CHAIRMAN APOSTOLAKIS: That's right.
MR. MAYS: So, I mean, I want to I think
it comes to my mind at this point to recall a
principle that Mike and I have talked about long and
often, with respect to this and other work, and we
have worked by that principle from the beginning, and
that's our principle, is progress, not perfection.
The idea is, we want to make incremental improvements
where we can, and we are not going to worry about the
fact that we don't have perfection either in what we
started with or what we end up with, because we don't
want to end up with what I loosely refer to as the
source term problem, where we start out with TID
14844, which was several people gathered around the
table thinking what they thought best, and then no
matter how much subsequent technical analysis gets
done, it becomes difficult to change, because the
other thing was already there.
So, we are trying to say, what we have is
what we started with. The ROP is there. I'm not here
to say whether the ROP is perfect or not, that's not
my job. My job here is to try to address things that
could make the ROP better, and that's the tone in
which we are trying to do this.
MR. JOHNSON: Yes. I would just add to
that two years ago we had what we had with respect to
the performance indicators, and we picked for targets
of opportunity for which we had data, and we set
thresholds as best we possibly could, and that
included for the green/white threshold that 95
percentile breakout for performance indicators.
Keep in mind that performance indicators,
some of the performance indicators were new, and were
in areas where you didn't have couldn't apply a
risk model, for example, the Security Equipment
Performance Index PI was a new PI, and there's no way
you can risk inform that, if you will.
So, but remember, in the broad context of
the ROP the green/while threshold is meant to be
indicative of an area where we need to go out and do
some additional inspection to look. So remember, don't
view the performance indicators in a vacuum. They are
a piece of an entire program, by which we provide
oversight on licensee performance.
MR. BOYCE: To try and get us back on
track, I was on the last bullet here, and I think
we've covered all the points, with the possible
exception of the last one, and that is, is that one of
the significant comments that we heard early on from
industry was, is that the risk-based PI program does
represent a large increase in the number of
performance indicators. And, any time you increase
the number of performance indicators you have the
opportunity for an increased opportunity for one of
the performance indicators to exceed a threshold.
Again, using the green/white threshold, if we go into
the white or above regulatory action is mandated under
our current ROP. And so, industry's comment was,
there's definitely an increased chance to regulatory
So, one of the things that we would
consider, if we were going to move forward with all of
the risk-based PIs, would be to, perhaps, modify the
algorithms on our Action Matrix for changing columns
from the licensee response column to the regulatory
response column, or other columns of the Action
Matrix. But, that is not, again, I don't to establish
premature expectations, that is not our current plan,
and I think Steve is looking at ways to, perhaps,
combine the performance indicators, so that there
would be fewer numbers, and I think you are going to
hear more about that.
But, we did want to say that we would
consider that sort of approach if necessary.
CHAIRMAN APOSTOLAKIS: Well oh, I'm
sorry, go ahead.
DOCTOR KRESS: If changing over from one
color to another across the threshold represents a
delta CDF, for example, then I don't see how I
mean, you have a total delta CDF you don't want to
exceed, you know, in your matrix, I don't see how
having more PIs changes that. If you set the change
for each one of the PIs to be a certain delta CDF or
related to it, then it doesn't matter
MR. MAYS: Actually, Tom, I think you are
correct, the issue would be, was the change in
performance reflected through the PIs or was the
change in performance reflected through inspection
finding, your point being that if you have had a
change in performance it should be reflected in one or
the other, and that change is the same regardless of
how it got found.
DOCTOR KRESS: Yes.
MR. MAYS: that's true, but I think it is
true also that there is what I would call an optics
issue, which is, if you have more direct PIs you have
maybe a faster responding optics with respect to the
fact that something has changed, and the Action Matrix
was set up on the basis of the limited number of PIs
that you have. So, the issue here was, the Action
Matrix was a little was defined in light of those
numbers of PIs, and so, therefore, it might be
something that would have to get looked at.
I think it's pretty clear that the more
PIs you have, the more opportunities you have to cross
a particular threshold, and that was basically the
concern that industry raised for us and
DOCTOR KRESS: Well, you know, that's what
I viewed as the good part, about adding more PI.
MR. MAYS: Well, it's the double-edged
DOCTOR KRESS: Right.
MR. MAYS: If you have more PIs you have
more opportunities to look green, but if you have more
PIs there's also another opportunity to have not
green, and the question is, how much of a value, I
guess, would be more greens as compared to more non-
greens, and I can't answer that question.
CHAIRMAN APOSTOLAKIS: That's an issue I
wanted to raise when I read the report. In Chapter 2
of the main report, January, 2001, there are four
steps that are listed in the RBPI development, assess
the potential risk input of degraded performance,
obtain performance data for risk-significant equipment
related elements, identify indicators capable of
detecting performance changes, and identify
performance thresholds consistent and so on.
It seems to me there is a major
consideration missing here, which is related to this
concern that we just discussed. When you come up with
a new indicator, shouldn't you be asking at some
point, is this information redundant with respect to
what I already have?
DOCTOR KRESS: That is the key question.
CHAIRMAN APOSTOLAKIS: See, you have to
constrain the number. If I look at these four and I
didn't know any better, because I'm sure you guys
thinks about it, but maybe you didn't state it, but if
I look at these four it's an open-ended process,
because it doesn't, at any moment trying to limit the
additional information that I'm getting from the RBPI,
and what's worse, at no point do you go back to the
baseline inspection and say, well, I've added this
performance indicator, therefore, I don't need to do
this now in the baseline inspection, and that I think
explains the concern from the licensees. All they see
is more PIs without anything else changing.
MR. BOYCE: I was going to say, you do
sound like an industry stakeholder at this point,
which explains the tone, perhaps, in that original
CHAIRMAN APOSTOLAKIS: But, there should be
something to limit the number, though. There should
be a tradeoff somewhere.
MR. MAYS: George, you're correct, and that
process is what the ROP change process is designed to
do. Our task from the RBPI development process was to
go and determine what was potentially possible to have
more direct measurement and indication of as
performance indicators for the ROP, in light of the
areas that NRR asked us to go look at. And, you are
right, this process does not limit the number.
However, we recognized, in coming up with
the number that we had, that that was a potential
issue, and NRR has recognized it, and the industry has
recognized it, and I think the judgment as to are more
indicators better, and are more indicators of value,
is something that the ROP change process has been
explicitly designed to try to answer.
So, I think that is something that we
expect will get dealt with through the ROP change
process as NRR looks at what we have technically
developed and determines whether or not it makes sense
CHAIRMAN APOSTOLAKIS: Could you add a step
like that to the
MR. MAYS: My point, George, is, that's
their step, that's not our step. Our step is to do
the feasibility to see what's technically feasible to
CHAIRMAN APOSTOLAKIS: Ah, okay.
MR. MAYS: Their job is to determine, once
we've got that technically feasible product, whether
or not it makes additional benefit to the process, and
that's what the ROP change process is designed to do.
MR. JOHNSON: And, if I can add, I just
checked on the way over, George, this morning, and, in
fact, the Inspection Manual chapter that provides that
change process is Inspection Manual Chapter 0608, and
it was issued earlier this week. It's available on
the internal web, and it will be available shortly on
the external web, and it provides for considerations
of the very things that you mentioned, does it add new
data, new information, what, in fact, changes ought we
be considering with respect to the baseline as a
result of those changes.
CHAIRMAN APOSTOLAKIS: But, if I look at
the beautiful figures that Research has developed,
like Figure 2.1, RBPI development process, where the
diamonds say do statistics accumulate quickly enough
to support timely plant-specific evaluation? Yes/No.
Timely quantification. Yes/No. There should be a
diamond somewhere there that says is this additional
information useful? Yes/No. Has it already been
covered? See, it falls naturally there, I think.
Now, whether somebody else does it is a different
story, but I think this diagram can be the basis for
evaluating this additional information, and then
addressing the licensee concern, which I think is
legitimate the way we are doing it. I mean, we are
just adding things.
MR. JOHNSON: Yes, I guess the answer we
are trying to give you is that those considerations
are already built into the process, the change
process. It has diamonds with Yes/No and you advance
you don't advance based on the answers to the kinds
of questions that you are asking, and we see that.
Again, what Steve has said is, Research's effort has
been the feasibility study, based on the results of
that feasibility study as we go forward and take
candidate risk-based PIs, we run them through that
process, before implementation we have answered all of
DOCTOR KRESS: George?
CHAIRMAN APOSTOLAKIS: Yes. I'm getting a
DOCTOR KRESS: Unless degraded performance
manifests itself as a uniform change across, say,
systems and components that are risk significant, so
that when you have degraded performance they all
degrade to some extent, then I don't see how you can
think that there might be redundancy or things covered
already, because all they are adding is risk
significant components and systems.
Now, if they add systems they could be
redundancy to components, of course, that would be the
only place I would worry about it, but otherwise,
unless you are
CHAIRMAN APOSTOLAKIS: For the initiating
events what you are saying might be more value.
DOCTOR KRESS: I think it's true for
reliability and availability also.
CHAIRMAN APOSTOLAKIS: For the mitigating
systems, I'm not so sure, but even for the initiating
events, it's not just a redundancy of information, but
maybe you can consider like I think they are
already doing that, things for which you do have some
records, and others that are really so rare that you
can't build a construct a performance indicator,
and maybe if you look at this class, for example, you
can pick one that would be more or less
representative, rather than having all of them. I
mean, you can bring additional considerations into
this to try to limit the number of
MR. MAYS: George, I think you've been
reading the script again. If we can get to the point
of the things that we've tried to do to address this
issue of the number of indicators, what we've referred
to as an alternative approach, I think you are going
to see a lot of these questions or issues dealt with.
We have looked at things of that nature,
and so I'll make the suggestion that maybe we get into
the meat of it, and you'll see where that comes out.
DOCTOR KRESS: Well, let me ask one other
question before we get there, is when you developed
your thresholds, for example, your delta CDF related
thresholds, you did them one component at a time.
Now, somewhere along the line you may end up with a
number of these things degrading.
MR. MAYS: You're reading the script again.
DOCTOR KRESS: Is that in the script
MR. MAYS: That's in the script.
DOCTOR KRESS: Okay, well, I'll just wait.
CHAIRMAN APOSTOLAKIS: Is this the last
time we are talking about the NRR reaction today?
MR. MAYS: Unless something else comes up
as we discuss the implementation.
CHAIRMAN APOSTOLAKIS: I want to ask a
question on the memo. Is that appropriate at this
MR. MAYS: You can ask anything you'd like,
CHAIRMAN APOSTOLAKIS: On page 7, there's
something I don't understand, but it appears to be
related to something that Doctor Kress and I have been
discussing over the years, it has to do with shutdown
PIs and it says, "Using the current process of
comparing time and risk-significant configuration to
a year does not seem appropriate for shutdown
conditions, since the entire outage may not be a
significant time interval compared to a year," 14 days
to 365. "As a suggestion ...," this is now what I
don't understand, "... perhaps, time spent in the
risk-significant condition as a percentage of plant
outage time would be a way of quantifying this risk
significance." Can you explain that a little bit,
what the rationale is, percentage of plant outage
MR. BOYCE: I'm not sure I can without
reading the memo. I can only offer to you that the
way that memo has developed, we sent around the Phase
1, draft Phase 1 risk-based PI report to several of
our technical branches, and we brought comments
together in that one memo. So, I cannot recall the
specifics of why that particular comment was written
the way it was.
CHAIRMAN APOSTOLAKIS: And, it says
DOCTOR KRESS: It sure sounds like a bad
idea, doesn't it?
CHAIRMAN APOSTOLAKIS: Yeah. First of all,
I'm trying to understand it. "Using that measure,
shorter outages would result in higher risk
significance." Now, that I just I would like to
understand a little better what the rationale for that
is, but, I mean, if you can't answer now, you can't
MR. BOYCE: I can't answer it definitively
CHAIRMAN APOSTOLAKIS: Is there any way we
can find out, Mr. Markley?
MR. HAMZEHEE: George, I think in general
what they are trying to say is that if you shorten the
outage you end up doing a lot of maintenance
activities at the same time, during a short period.
As a result, you have more equipment out of service,
and if something goes wrong then the availability of
your safety systems are limited.
CHAIRMAN APOSTOLAKIS: Yes, but this is a
very qualitative statement that, you know, somebody
can come back and say, gee, I'm using my PRA, I'm
using OREM (phonetic), Sentinel (phonetic), and all
these things, and I'm controlling on these things, so
how can you, you know, speculate? And also, this
becomes more specific, it says, "We can compare the
time spent in the risk-significant condition as a
percentage of plant outage time," in other words the
plant outage time has some magic to it.
MR. MARKLEY: The k heat load would be the
primary thing if they go into reduced inventory, even
though they have done a lot of maintenance on line.
MR. MAYS: Let me suggest that rather than
us speculate, that if you would like to get an answer
to that we will try to determine who made the comment
and try to get something out to you.
DOCTOR BONACA: Yes, I'd rather differ, as
the comparing time and risk-significant configuration
to total outage time, in that sense if you are
attempting to shrink the whole outage time by, for
example, staying a longer time in a risk-significant
configuration, okay, versus staying with a longer
outage time, total outage, by reducing the time in
risk configuration to a shorter time, that's the
comparator I see there.
MR. MAYS: I read that as a more general
concern, quite frankly, George.
DOCTOR BONACA: Assume that you have we
are going to go through some configurations, some are
riskier than others, and you may find that you may be
able to shorten the whole outage by staying a longer
time into a risk-significant configuration. Okay?
That's the concept, it seems to me.
CHAIRMAN APOSTOLAKIS: Perhaps, what we can
do is, can we ask NRR to send us a little memo
explaining this? Is that
MR. JOHNSON: Yes. I would almost suggest,
if we could come over and I mean, I'm not sure what
your schedule is like, but we would certainly your
question is a good one, and we certainly look forward
to trying to provide
CHAIRMAN APOSTOLAKIS: No, we can address
it at the full committee, you can address it at the
full committee meeting. It can be an item to you
have plenty of time until then.
MR. JOHNSON: Sure. Sure. When is the
full committee scheduled to meet, I'm sorry?
MR. MAYS: I believe we're on Friday on the
CHAIRMAN APOSTOLAKIS: The first week of
MR. MAYS: The first week of May, is it the
CHAIRMAN APOSTOLAKIS: So, you have two
weeks at least.
MR. JOHNSON: Yeah, let us come back at
that time with an answer to your specific question.
CHAIRMAN APOSTOLAKIS: Okay.
MR. MAYS: And, George, the way I read that
comment was a little less specific than you did. The
way I read that comment was as follow, as licensees go
to shorter and shorter outage times, a greater
percentage of their outage time is spent in high
relative to decay heat (phonetic) scenarios, and some
percentage of their time is spent more in mid-loop
operations, and the concern was, is that constitutes
a higher risk situation. And, the concern was whether
or not the indicators, as we've proposed in the RBPIs,
would be capable of dealing with that particular
Now, I think they do. I viewed that as a
challenge to me to get back to the commenter and
explain to them how these RBPIs will deal with the
fact that if they go to shorter and shorter outages,
and they involved greater risk scenarios, that these
would be capable of detecting them. That was the way
I took that comment.
And so, I think it's covered, but that's
part of the process we'll have to do to get back with
the people we've received comments on, as we go
through to make the final report.
CHAIRMAN APOSTOLAKIS: Okay.
DOCTOR KRESS: Certainly, it seems to me
like the appropriate thing is just what you've done,
and that's time in risk-significant configurations.
MR. MAYS: I think it addresses that
comment, but it wasn't clear to that person making the
comment that it does.
DOCTOR KRESS: Yeah, that ought to be the
appropriate way to look at it.
DOCTOR BONACA: Yes, I think here central
is the statement above in the title that says,
"Licensees are currently performing so refueling
outages are of very short durations," and that's the
focus of that. You can be, you know, more capable of
going shorter, but
DOCTOR KRESS: That ought to be covered
with what they've got.
CHAIRMAN APOSTOLAKIS: I understand it
qualitatively, but I think this goes beyond that, it
actually tells you how to do it, and I'm trying to see
what the implications would be to Regulatory Guide
1174, because you have been arguing for a long time
that it's the average over the year, and these guys
seem to be going away from that. So, I'd like to have
some further discussion. It's not just in this
context, okay, but this is something that has been of
concern to Doctor Kress and me for a while now.
MR. MAYS: Well, let me suggest that the
context that would be most appropriate for you is for
us to go back and discuss this with the person who
made the comment, and then when we have come up with
a solution, present what the solution is to you and if
you agree with it then it doesn't matter what the
CHAIRMAN APOSTOLAKIS: That's fine.
Okay, so I don't know why Tom took so long
to finish just the
MR. BOYCE: I apologize for that.
CHAIRMAN APOSTOLAKIS: Apologies accepted.
MR. MAYS: Tom has difficulty not talking
a lot, and he really is
CHAIRMAN APOSTOLAKIS: So, we'll go back to
MR. BOYCE: Steve made the comment we
shouldn't send him comments.
CHAIRMAN APOSTOLAKIS: Okay, Steve.
MR. MAYS: Okay.
The rest of what we are going to present
today is primarily the results of our stuff. Mike and
Tom will be sitting over here at the side if there's
any other questions.
So, what I suggest we do here, if it's all
right with you, the first portion of this is
discussions of the potential benefits before we get
into the summary of the thing, so if you want to do
those first and then I didn't know what time you
wanted to take your first break. Okay, so let's go
through the benefits first.
What we have outlined in this report is
some of the things that we think are the benefits of
RBPIs, and the first one which answers part of the
question you raised, George, is why we even want to do
this. Well, one of the reasons is, we get a much
broader sample of risk performance with this set of
indicators than we do with the current ROP, and they
are a more objective indication because they are more
directly tied to plant-specific performance and with
a relationship to the plant-specific thresholds.
So, we believe that's a positive, that's
one of those progress versus perfection things, that's
one of those potential benefits that we think this
Also, years ago, NEI submitted a document,
a white paper, to us, NEI 96-04, which was their paper
on risk-based and performance-based regulation, and
they wanted us to move in the direction where we had,
as just quoted here, a regulatory approach that more
directly considered operating experience and the risk
implications of it, and performance-based process
where we had measurables, and objective criteria, and
specified reactor or, specified activities that the
NRC would take and flexibility for the licensee as
long as they were performing in an appropriate band.
Well, I think the ROP process reflects those general
principles, and the RBPIs are an example of a more
direct approach to applying operating experience and
probabilistic safety assessment judgments as to how we
would go about doing that.
DOCTOR KRESS: Steve, I think the word
"sample" within that dot is a really key word.
MR. MAYS: That's an important word.
DOCTOR KRESS: Because you are not
measuring the full performance always, you are taking
MR. MAYS: That's correct.
DOCTOR KRESS: And, you are going to infer
from that what the full performance is, and I think
that's a key concept in this whole thing.
MR. MAYS: I agree, that is a key concept.
The issue that's part of built into the Reactor
Oversight Process is that the indicators will be a
sample of performance, and the inspections will be the
process by which we go out and sample the rest of the
performance, as it relates to meeting cornerstone
objectives. So, again, this is a balance of how much
of your sampling you want to spend in the PIs, how
much of your sampling do you want to spend in
inspection, and remember, a key part of this Reactor
Oversight Process is not that the NRC does all of the
sampling, it's that the licensees do the sampling,
that their problem identification and corrective
action programs are the key behind all this, that they
are continuously sampling and looking for things, and
we have a smaller subset that we look at to provide us
with the assurance that they are doing their job
right. So, that's an important point, I think, to be
In doing the sample with the RBPIs that we
proposed, we've got more systems and more components
covered by more objective and more risk-informed
methods than the current ROP has. And, in the
indicator space, we have some indicators that go
across system boundaries and across the breadth of the
plant, and we believe that's an important piece
because one of the issues earlier raised was what
about crosscutting issues? Well, what if I have my
maintenance program degrading and I just don't happen
to see it in my diesel generator or my HPI indicator,
how will I know that my plant is getting worse? Well,
by having some of these indicators that go across
systems, we think that might help address some of
those issues from an indicator standpoint. The rest
of it has to be addressed through inspection.
CHAIRMAN APOSTOLAKIS: But, you are saying,
on page A-25, "Currently, there is no established
method of identifying changes in operator performance
and then feeding this information back into the SPAR
models. As a result, equipment performance is the
only mitigating system out there that will be
evaluated in this analysis." Are you saying there
that the crosscutting issue of safety conscious
environment, and the corrective action program, cannot
have performance indicators, we have to do something
else about them?
MR. MAYS: I'm saying I don't have anything
readily available now to do it. I'm not saying it's
impossible to develop it, but I'm saying I don't have
that capability right now. The capability I have
right now is to reflect whatever operator performance,
with respect to safety culture, with respect to
maintenance program, as to how they manifest
themselves in respect to the availability and
reliability of the equipment. So, I can't directly go
out right now and measure the safety culture at the
plant, but I can go out and measure whether the safety
culture of the plant has had an impact on the
availability and reliability and the frequency of
CHAIRMAN APOSTOLAKIS: But, I thought
equipment performance was taken as a separate
attribute from the human performance. In other words,
if it's a valve, and it is left inadvertently closed,
would that be part of the indicator for the valve?
MR. MAYS: Yes.
CHAIRMAN APOSTOLAKIS: Because even though
it was not a fault of the valve itself?
MR. MAYS: Correct.
MR. HAMZEHEE: But, it wasn't available.
CHAIRMAN APOSTOLAKIS: Huh?
MR. HAMZEHEE: But, it wasn't available.
CHAIRMAN APOSTOLAKIS: It was unavailable.
MR. HAMZEHEE: It's reflected in the
unavailability of that equipment.
CHAIRMAN APOSTOLAKIS: And, you will keep
track of the fact that it was a human error?
MR. HAMZEHEE: The cause would show, yes.
MR. MAYS: Well, the RBPIs would not
reflect the fact that it was a human error. The basic
data that was going into the RBPIs would be available
to us, so that if we determined that somebody's
performance was requiring additional regulatory
attention, we could go back and look at the
information and say what was causing this to be a
problem, and then use that as part of our guidance for
how we go and look at the plant.
The issue that we were raising in that
particular point was that we don't have direct human
performance measures that we are going to have
CHAIRMAN APOSTOLAKIS: So, these then
crosscutting issues should be part of the baseline
MR. MAYS: They are.
CHAIRMAN APOSTOLAKIS: Okay.
MR. MAYS: And, this would be a case where
we would have more direct objective indicators of some
of the impacts of that.
CHAIRMAN APOSTOLAKIS: I thought they were
DOCTOR KRESS: I didn't think they were.
MR. MAYS: Well, the crosscutting issues
no, the crosscutting issues are dealt with in the ROP
through the problem identification and resolution
inspections, to determine whether or not the plant has
an appropriate process by which they can manage those
kinds of issues.
CHAIRMAN APOSTOLAKIS: Well, is that true,
Mike? I don't remember. Oh, it's not that he's
MR. JOHNSON: I'm sure it was true,
although I've got to confess I was talking. I didn't
hear the total comment.
MR. MAYS: The additional benefits that we
alluded to earlier has to do with the fact that we
have a better understanding of plant-specific
implications using these than we necessarily had with
the current ROP. Our thresholds are set on the basis
of plant-specific models. We don't average diverse
systems together, which can potentially mask the
contribution. For example, in the ROP, the turbine-
drive pump trains and the motor-drive pump trains are
AFW, their unavailability is averaged, and that's the
value that's used in the PI. Well, turbine or diesel-
driven pumps have different risk significance than
motor-driven pumps because of the station blackout
risk issue, and so the RBPIs that we proposed allow us
to deal with that.
The other thing that I think has come up
on a couple of occasions in the current ROP that has
been dealt with in the RBPIs is whether the failures
affecting the reliability and availability indicator
that you might have, whether they are based on the
risk-significant functions or whether they are based
on design basis functions. The example that comes to
mind was the, I believe it was Quad Cities had a case
where they ran their once a cycle test of their HPCI
system to see it it would automatically actuate, and,
in fact, it wouldn't. There was a problem with the
automatic circuitry to start the HCPI system.
Now, over the period of the cycle, they
had been manually starting the system every month or
quarter or something like that, and it was working
just fine. So, what happened was, they determined
that they had a problem with the automatic feature for
this system, and the fact that they had not tested it
since the last outage meant that they had nine months
of fault exposure time to put into the indicator.
Well, that indicator had nine months of fault exposure
time, which only represents that it wouldn't have
performed its automatic start capability, while it's
manual capability was not degraded at all.
And, from a risk perspective, having the
manual ability to start HPCI is success, so one of the
things we've done in the RBPI program is to deal with
the difference between auto and manual and design-
basis requirements versus risk-significant
requirements for the equipment to operate.
We've also had a different way of treating
fault exposure time than was in the current ROP, which
we believe is more consistently accounted for and is
more consistent with the way risk assessments are
done. The issue there having to do with the fact that
in the current ROP there are no reliability indicators
per se. Fault exposure time was included in the
availability indicator as a sort of surrogate for
having a reliability indicator, and because of the
relatively short time period under which the
availability is gathered, and the fact that the fault
exposure time every time you do have one of these
failures can be fairly long depending on its nature,
you have a false positive/false negative problem which
goes back to the old thing that Hal Lewis always
talked about, trigger values. The RBPIs don't have
that same problem because we classify the failures as
either demand-related failures or not, and for those
we use a probability calculation and distribution for
reliability rather than use the fault exposure time.
For fault exposure times associated with discovered
events, for which there was no demand, those go into
the unavailability in the RBPIs. So, we have a more
consistent way of dealing with that, which we believe
tends to reduce the problems that were currently being
experienced in the RBPIs or in the oversight process
with fault exposure time.
DOCTOR KRESS: When you determine
unavailability, is it true that you count into that
unavailability time the time spent testing a piece of
MR. MAYS: If it's out of service and not
capable of being used while that test is going on,
DOCTOR KRESS: I personally think that's a
mistake to do that, but we can discuss it later. It
does a lot of it has a lot of negative aspects to
counting that in there, one of which is, when they do
this testing, they are on a higher alert and the
operator error in doing some compensatory measure is
probably much less than it would be, so the risk is
not the same as it would be if it were just
unavailable because it was not functioning correctly.
And, not only that, it gives a negative
incentive to not test as often.
MR. MAYS: Well, only if you are doing a
lot of testing to the point where it might reach a
threshold to contribute to your
DOCTOR KRESS: Of course
MR. MAYS: this is the classic issue
from the maintenance rule.
DOCTOR KRESS: I'm being too general
with this, but then
MR. MAYS: This is the classic issue from
the maintenance rule, the balance between the time you
spend in testing and maintenance and the impact on
reliability and risk. So, that's a problem, I haven't
resolved that problem, I'm just trying to be
consistent with the current approach.
Additional benefits that we have, this
process was designed so that the RBPIs would look
similar to the current performance indicators, that we
would have similar color scheme, they'd be amenable to
similar kind of presentations on the web site, and
they could be updated in a similar fashion that the
current process has.
One of the things we've also noted is that
these don't have to be implemented, it's not an all or
nothing deal. In other words, portions of these can
be implemented, some of them can come along later as
data, and availability, and quality become better, so
this is not an all or nothing deal.
DOCTOR KRESS: The nice thing about these
performance indicators that you have now is, you could
actually calculate a delta CDF. You could take the
set of performance indicators at some time and stick
them in a plant-specific model and get a delta CDF.
MR. MAYS: You are reading the script
again, Tom. That's correct. One of the things we had
as part of the Phase 2 work that we had originally
proposed was to look at how we might develop an
DOCTOR KRESS: That could be the integrated
MR. MAYS: So, that's part of what you are
going to see a little bit later on.
You were correct in stating earlier that
all of the indicators we have now in the report, and
the current Reactor Oversight Process indicators, are
all basically single variate sensitivity analysis on
a larger model.
DOCTOR KRESS: Right, but you could take
the whole shebang and put it in and calculate it.
MR. MAYS: The issue then is, are there
synergies among these things that would make them go
up, down, or sideways, if you had a more integrated
look. We'll talk some more about that as we get
The other thing I wanted to mention,
because this became a point of confusion with people
both internally and externally, that the RBPIs, while
we went back and did a lot of work looking at
statistical methods to determine what's the right time
intervals, what's the right method of calculating
these things, and what's the process for setting the
thresholds, that this isn't something dramatically
exotic. We are using off-the-shelf, readily-available
models. The analysis routines that we are planning to
use are fairly simple, and most of the data we've got
is from readily-available current databases, there's
not, with a couple exceptions, any new information
that really needs to be required to make this happen.
So, most of the stuff is fairly easy to get and to
So, we can get into some of the results
now. We talked about the four elements, George
brought them up earlier, about how we were going about
doing that. We wanted to look for areas where there
would be risk impact of performance if the plant was
degraded, find out if we could get data on that
information, make sure that if we did that we could
the tech changes in a timely manner, and then be
consistent with the 99-007 general rule process.
Now, what that means in a practical sense
is that, in order to do that you have to have three
things. You have to have a model that reasonably
reflects the risk at the plant. Now, the word I want
you to concentrate on there is reasonably. We were
talking about the progress, not the detection mode
before, what we have to have is a model that has some
fidelity to the risk at the plant, in order for us to
believe that we have something that goes on, is real.
Then we have to have some baseline performance to put
in that model in order to be able to say, this is our
starting point, and then we can vary the model off of
the baseline to determine what the impact is of
changes in the performance.
And, the last thing you have to have in
order to be successful at doing this is, you have to
have an ongoing source of performance data for
assessing the plant-specific performance. And, what
you'll see as we go through the rest of these is,
there were some cases where we had all three of those
things and we've made proposals, and some cases where
we didn't have them, and so, therefore, we weren't
able to do performance indicators on those areas.
CHAIRMAN APOSTOLAKIS: Should we take a
MR. MAYS: Sure, if you want to take a
break now, that's no problem.
CHAIRMAN APOSTOLAKIS: Until 10:00.
(Whereupon, at 9:44 a.m., a recess until
CHAIRMAN APOSTOLAKIS: Back into session.
Mr. Mays, continue, please.
MR. MAYS: Okay.
The first thing we are going to talk about
from the results of the RBPIs is the work we did in
the initiating event cornerstone, which was for full
power internal events. We used three data sources for
putting this stuff together, new Reg 5750, which was
the initiating event report which we did a couple
years ago and you've seen. We used the Sequence
Coding and Search System, which has the LER
information, which is the source of information about
plant trips, and the Monthly Operating Reports, which
gives us the critical hours information for the
plants. All those sources, by the way, are publicly
available, there's no issues with respect to
availability and quality of that stuff as far as
So, we went back and in going through the
process we just discussed we determined that there was
three RBPIs we could do for each plant, and the tables
are listed as to where they can be found in the main
report and the appendices. The important part about
here was how we came up with the calculations of the
frequencies. Now, the current ROP merely counts the
number of SCRAMS you have and goes on from there. We
were looking more at the classical PRA definition of
establishing a frequency which has distribution
associated with it, and so we were looking to see what
we could do in terms of prior distributions for
figuring these out, and we had three options that we
One was to start with, basically, a non-
informative prior, kind of a classical statistical
approach, how many failures did you have in how many
years, and that's your estimate.
The next thing we looked at was taking an
industry prior, which would be to say you would take
the distribution of the industry population and update
that with the plant-specific information.
CHAIRMAN APOSTOLAKIS: Can you tell me,
Steve, where you did all this stuff?
MR. MAYS: It's in Appendix A, I believe.
CHAIRMAN APOSTOLAKIS: Appendix A, I don't
recall seeing prior distributions. Maybe I missed it.
MR. MAYS: Just a second, let me find it.
MR. HAMZEHEE: Steve, Appendix F.
MR. MAYS: F?
MR. HAMZEHEE: Statistical Methods and
CHAIRMAN APOSTOLAKIS: Oh, F.
MR. HAMZEHEE: Yes.
CHAIRMAN APOSTOLAKIS: Okay, thanks.
MR. MAYS: So, and then the last one we
tried was one that you've seen before in reports that
we've given you on system and other studies on
constrained, non-informative prior.
CHAIRMAN APOSTOLAKIS: So, this appendix
will tell me how the choice of the interval
observation was made?
MR. MAYS: Yes, right.
CHAIRMAN APOSTOLAKIS: So, what is it,
between one and five years?
MR. MAYS: Well, that's the next bullet
down. Let me explain what we were doing. We tried
three different priors to see which one would give us
the best performance that we were looking for, in
terms of being able to give us timely indication, not
give us too many false positives or false negatives,
and to be amenable to being done with the ROP process.
So, as it turns out, we were looking at
the time intervals. What we wanted to do is take the
shortest time interval that would give us indication
of performance degradations for which we wouldn't have
a false positive or false negative rate that was
excessive. And, by a false positive rate, what I mean
is that there would be a significant chance that
performance hadn't degraded, but the way you calculate
it it would send you over the threshold.
Then, the false negative would be the
situation where if you had a significant degradation
in your performance that you would go over the period
of time that you were looking and wouldn't have enough
data collected to see the changes that occurred. So,
that's the simple basis of what we did.
So, when we looked at that for the
initiating events, we used one year as the time
interval for the category referred to as general
transients, that's trips, the plant trips, but the
safety systems needed for decay heat removal, for
activity control, that kind of thing, are not affected
by the trip itself, and we also came up with three
year intervals for loss of feedwater and loss of heat
sink events, which are trips that are a little more
complicated and the ability to remove decay heat is
impacted directly by the trip itself.
For other risk-significant initiators that
you typically find in PRAs, like losses of off-site
power, steam generator tube ruptures, small LOCAs and
other initiators, the problem we had there was that
the frequency of occurrence on a plant-specific basis
of those things was so infrequent that over a you'd
have to take more than a five-year period to be able
to even see that, and that didn't seem to be
consistent with the ROP philosophy, which was to go
back every year and see where the performance was
going so that we could see what we needed to do more
of with respect to those indicators.
CHAIRMAN APOSTOLAKIS: Well, that's a very
important point, though, and I must say that I haven't
really read Appendix F in detail, but I was doing my
own calculations and I used as an example Table A.1.4-
2C plant, two initiating event threshold summary. It
seems to me that the results of the aleatory part, the
randomness issue that we have addressed here, which is
something that the quality control people do, so we
have two thresholds here. One is green/white, which
is .8, right?
MR. MAYS: Which page are you on there,
CHAIRMAN APOSTOLAKIS: A-17, it's just
numbers on here, but A-17. So, we have the
green/white 8E-1, right, and the baseline was 6.8E-2,
right, the same table?
MR. MAYS: Yeah, why don't you just flip to
the next page in your presentation, that particular
chart is right here.
CHAIRMAN APOSTOLAKIS: Okay, fine, but if
the question is now, how long should the interval
be, so that the calculation of the rate will be
meaningful, and I would have some sort of conclusion
that I'm near the baseline or the white region, and
for the numbers here I calculated that to be about ten
years, which is really too long, as you just pointed
And, the problem is this, that because
these numbers are very low, if you see nothing, that
doesn't necessarily mean you are near the baseline,
you could be near the you could be in the white,
because it's .8. If your observation is one year
MR. MAYS: Well, in this case, loss of
feedwater is three years.
CHAIRMAN APOSTOLAKIS: Okay, the three
years I think we are beginning now to be a little
better, but I think the analysis maybe I should
send you a little memo with what I did so you can tell
me what I did wrong.
MR. MAYS: No problem. The transient
initiator we used one year, loss of feedwater, loss of
heat sink we used three years.
CHAIRMAN APOSTOLAKIS: It's three years,
MR. MAYS: And, when we got to that point,
you still have the possibility, because this is a
distribution, we are calculating a frequency and it
has distribution, but we are comparing the mean of
that distribution to a specific value for the
threshold. So, there's always the possibility of
false positive/false negative with that.
CHAIRMAN APOSTOLAKIS: What's new here, I
think, not new, but in PRAs typically we deal with
systemic uncertainty, the uncertainty, failures rates
and the initiating event frequency. Here you have to
worry about the aleatory part, too, because you are
talking about real occurrences. So, the fact that
they have seen none in the last two years is that due
to chance, and my rate of occurrence is, in fact,
high, but I just happened not to see it, or is it
because the rate is low, and that's the key question
that the quality control people are asking.
MR. MAYS: Right, and what we've done there
is, we've asked a slightly different question. We
didn't ask the question, is the mean what we are
calculating here, the "correct mean." What we are
saying in this situation is, if there was substantial
degraded performance, is a three-year period enough so
that we would be able to detect that it wasn't green
anymore, and the answer to that question is yes.
CHAIRMAN APOSTOLAKIS: With a certain
MR. MAYS: Right.
The other side of that coin is, okay,
suppose I do have a few events in a one, or two, or
three year period, does that necessarily mean that my
frequency has, you know, gone up.
CHAIRMAN APOSTOLAKIS: Yes.
MR. MAYS: And, what I'm saying is, the way
we've dealt with that is that we've dealt with that
issue, the problem I think you may be looking at is
the classical issue is, if my frequency is around .07
or so, then in three years can I get enough faults
where X over three years tells me something. That's
the problem you get when you use the classical
approach. What we've done instead is, we've said the
industry average is this number, .068, we used a
constrained non-informative prior, and we've used a
Bayesian update of that to get the new distribution.
So, what we don't have is, we don't have the same
amount of problems with either the inability to detect
changes with the classical approach from a false
negative standpoint, or a positive standpoint, and we
don't get the false negative problem we have when you
use the industry prior by itself, which means that you
have to have an awful lot of data at the plant to
overwhelm the industry prior.
So, the constrained non-informative prior
seems to be the middle ground between those two
extremes that works best, and that's what we chose to
use because it had the lower it had the performance
characteristics because it's a competing interest.
False positives and false negatives are competing
interests, so that's the way we did that.
MR. HAMZEHEE: I think, George, maybe when
you were doing your calculation you did not use any
CHAIRMAN APOSTOLAKIS: I didn't.
MR. HAMZEHEE: That's the reason.
MR. MAYS: That's the classical approach.
MR. HAMZEHEE: So, you just used a direct
number and you get ten or 20 years sometimes to get a
MR. MAYS: Right.
CHAIRMAN APOSTOLAKIS: Well, actually, for
the one standard deviation the interval is only 2-1/2,
so it's not bad, 2-1/2 years, it's reasonable.
MR. MAYS: Yes, as it turns out
CHAIRMAN APOSTOLAKIS: But, still, though,
there is a I mean, it's only one standard
deviation, so the probability that I'm wrong is not
MR. HAMZEHEE: Oh, yeah.
CHAIRMAN APOSTOLAKIS: But, I'm going to
read Appendix F, so at the full committee meeting
we'll have a more meaningful discussion.
DOCTOR KRESS: What is your rationale,
justification, for using the industry distribution as
CHAIRMAN APOSTOLAKIS: Yeah.
DOCTOR KRESS: Do you think that really has
the technical justification?
MR. MAYS: I think it does have a technical
justification. The issue there becomes one of and
this is a standard PRA issue that goes on you have
a limited number of a limited amount of data at any
one particular plant, and you have to go a long time
to collect data only at that plant. And, the example
I use for people is, go to Atlantic City, do I need to
go to every table on the roulette thing at every place
in Atlantic City and take infinite data on each one to
know something about their performance, or can I take
some data over a group and then go back to any
particular table and monitor its performance relative
to the group to see if it's different, and that's
basically the approach that we're taking here.
There's a ceratin amount of time that you
can take your sample for, to get enough information to
do what you need to do.
DOCTOR KRESS: I understand that
constraint, but I still don't believe
CHAIRMAN APOSTOLAKIS: I guess the
rationale is, my plant could be any one of these,
okay, and the plant-to-plant variability gives me the
fraction of plants that have this particular value.
It could be any one.
DOCTOR KRESS: Your assumption, though, is
that that distribution basically applies to the
distribution at that plant.
CHAIRMAN APOSTOLAKIS: Yeah, that it's one
MR. MAYS: No, not quite, Tom. If you go
out and you were to calculate, like we did in new Reg
5750, what the frequency was for either PWRs, or BWRs,
or the population of plants, when we did that
calculation we put a distribution on that that
represented the plant-to-plant variability in the
population. What we are using here is not that
distribution itself, that would be the industry prior
distribution, and that tends to give you a problem in
that you can have significant degradation and it takes
you a long time for your plant-specific performance to
overwhelm that initial prior.
What we did instead was, we took that
industry distribution, we took the mean out of that
distribution, and then we constructed a constrained
non-informative prior where we diffused the
distribution, so that what you see when you do the
update is the impact more of the plant performance
than of the industry performance, and that helps us
resolve, I think, the issue you are talking about.
DOCTOR KRESS: Yes, I think that would
help. I still think there's a problem with choosing
that mean also. I'll have to read it a little more
MR. MAYS: We checked in the if you'll
recall, the 5750 to determine whether or not we were
seeing plant-to-plant variability on those things, and
so there's a means of being able to deal with that.
DOCTOR KRESS: Well, you know, eventually,
though, you keep updating and the problem will go
MR. MAYS: Correct.
DOCTOR KRESS: Eventually.
MR. MAYS: Given enough time.
DOCTOR KRESS: Given enough time, yeah.
MR. MAYS: So, that's what I was going to
do next, was turn to this page here with
CHAIRMAN APOSTOLAKIS: And, this 95th
percentile column is explained somewhere in the
MR. MAYS: Yes, that's what the value for
the threshold would be if you took the industry prior
CHAIRMAN APOSTOLAKIS: Oh.
MR. MAYS: set the 95th percentile on
DOCTOR KRESS: Just like they did in the
CHAIRMAN APOSTOLAKIS: So, the green/white
is on the second table .8, and the industry average
95th percentile would be .2, am I correct?
MR. MAYS: Correct.
CHAIRMAN APOSTOLAKIS: The industry would
be .2, l8, so it's higher?
MR. MAYS: In some cases.
CHAIRMAN APOSTOLAKIS: Plant-specific is
MR. MAYS: In some cases, in some cases
it's higher, and in some cases it's significantly
lower. It depends on how you go. There were examples
where that would happen, less so with the initiating
events, but more so with the availability and
CHAIRMAN APOSTOLAKIS: It should be lower,
though, should it not, as a rule?
MR. HAMZEHEE: Well, usually yes, because
you are talking about 95 percent.
MR. MAYS: Just a second, George, I think
we may be making a difference. The value in the 95th
percentile column is the value that corresponds to the
95th percentile of the distribution.
CHAIRMAN APOSTOLAKIS: The industry.
MR. MAYS: Of the industry.
CHAIRMAN APOSTOLAKIS: So, that should be
higher than 95 percent of the plants, right?
MR. MAYS: Correct.
CHAIRMAN APOSTOLAKIS: Right.
MR. MAYS: Now, what we are saying is, for
this particular plant, and remember the baseline was
.07, the 95th percentile in this case was .2.
CHAIRMAN APOSTOLAKIS: Uh-huh.
MR. MAYS: All right. We are saying that
the risk contribution from changing from .68 to .8
gives us the delta risk increment, whereas the 95th
percentile just tells you how much it varied among the
CHAIRMAN APOSTOLAKIS: Oh, oh.
MR. MAYS: There's no direct relationship
between those two. We were showing where you might
set the threshold if you used the 95th percentile
approach, which is deviation from normal performance,
versus a risk threshold approach.
CHAIRMAN APOSTOLAKIS: So, what they should
really be comparing is the first two columns, and
there the baseline is lower.
MR. MAYS: Correct.
CHAIRMAN APOSTOLAKIS: Okay.
MR. MAYS: And, what we found was, is that
sometimes, in certain cases, as we tried to apply this
concept uniformly through the plants, sometimes you
would find cases where you wouldn't exceed the
green/white threshold until you were already up in the
CHAIRMAN APOSTOLAKIS: Yes.
MR. MAYS: And, we said, that doesn't seem
to be a smart thing for us to do.
CHAIRMAN APOSTOLAKIS: So, how many plants
do you expect to see with 67 transients a year?
MR. MAYS: None.
MR. HAMZEHEE: That's just a number.
MR. MAYS: None, the point is, and this
goes back to Tom's earlier point, what we have
MR. HAMZEHEE: That was mine.
MR. MAYS: or your point, or whatever,
is that you have a single point variance analysis.
Now, what that tells you is that if everything else in
the plant were to stay the same, except for this
input, how high would it have to go to get me to an
increase in core damage frequency of E-4.
DOCTOR KRESS: But, you are never going to
see that. If you get that bad
MR. MAYS: The realities of I think
everybody will agree the realities are that other
things will go wrong before you get to that point, and
we'll find something and be able to deal with it
before it gets there.
CHAIRMAN APOSTOLAKIS: In other words, if
I go to the Action Matrix I will see enough whites and
greens, whites, way before I see any reds, unless it's
some sort of a major disaster.
MR. MAYS: Well, you know, I'm saying, I'm
not sure how anybody engineering-wise would be able to
trip their plant ten or 15 times a year without having
other problems in the plant that would manifest
CHAIRMAN APOSTOLAKIS: Without having the
DOCTOR KRESS: That's one reason I question
the usefulness of that whole problem.
MR. MAYS: And, I understand that, that's
always going to be the case when you have, risk is a
function of multiple variables.
DOCTOR KRESS: Yes, absolutely.
MR. MAYS: And, if you have indicators that
are single variable sensitivity analysis, you always
have the issue of, is it realistic that this is the
only thing that will change?
CHAIRMAN APOSTOLAKIS: Wait a minute, now.
Isn't that dependent also on the baseline core damage
MR. MAYS: Absolutely.
CHAIRMAN APOSTOLAKIS: For plants that are
already I mean, 19 units that are above, then you
shouldn't expect 67 to be in the red.
DOCTOR KRESS: No, you might
CHAIRMAN APOSTOLAKIS: In fact
DOCTOR KRESS: yeah, but
CHAIRMAN APOSTOLAKIS: as it should be.
DOCTOR KRESS: the question is, I'm not
sure where you see that reflected in the thresholds,
because the thresholds don't use the absolute value in
them. That's another thing that bothers me.
MR. HAMZEHEE: No, they use the impact on
MR. MAYS: Right.
DOCTOR KRESS: It's the delta, they just
use the delta.
MR. HAMZEHEE: Based on the delta CDF, you
set the value.
CHAIRMAN APOSTOLAKIS: Yeah, but if you are
DOCTOR KRESS: It doesn't matter.
CHAIRMAN APOSTOLAKIS: It doesn't matter?
MR. MAYS: Not quite.
DOCTOR KRESS: And, that bothers me a
MR. HAMZEHEE: Well, but it shows the
importance of that.
MR. MAYS: Not quite.
CHAIRMAN APOSTOLAKIS: It does not, do you
agree with that?
MR. MAYS: Not quite. It depends on all
the other things that are in the model together. This
is the issue we are raising in the first place. You
have some baseline core damage frequency, depending on
the model of your plant.
DOCTOR KRESS: Yes.
MR. MAYS: And, depending on that, and the
relationship between the initiator frequency or the
diesel generator reliability, or whatever else is in
your model, you can vary that, and if you start at a
lower baseline you have to have greater changes in
order to get to a E-4 delta CDF. However, if you start
with a E-4 delta CDF you still have to have a certain
amount of change to go to 2E-4, which is what this
threshold would be measuring. So, this threshold
measures change in the CDF of the plant, it does not
measure directly the total absolute CDF. You can go
back and figure it out if you wanted to, but that's
another issue for the integrated indicator, which was
the thing we talked about before.
MR. HAMZEHEE: It also shows for that
specific plant that general transient by itself is not
very risk significant. In other words, you are never
going to change your CDF by greater than 1-4 unless
you go above 67 trips per year.
MR. MAYS: Yeah, that's the other thing it
DOCTOR KRESS: Yes, and that's a
significant piece, an incite, I think. But, you know,
we are speaking in general when we talk about it, even
the other PIs, not just this one, that it seems like
the absolute value ought to be reflected in there
somewhere, and I don't think it really is.
CHAIRMAN APOSTOLAKIS: Somehow.
MR. MAYS: We chose as part of the ROP
philosophy that what we were going to do was, we
started with the basic assumption that the design and
operation of the plants was basically safe, and then
our job was to be able to detect changes in
performance in the plants that might be more risk
significant, so that we could engage them. So, that
philosophy is what determines this.
DOCTOR KRESS: Yeah, but you can turn that
around a little bit.
CHAIRMAN APOSTOLAKIS: That's a very good
DOCTOR KRESS: Yeah, but you can turn it
around and say there are some plants that are not just
basically safe, but, really, really good risk status.
CHAIRMAN APOSTOLAKIS: So, you are
DOCTOR KRESS: And, you are penalizing
MR. MAYS: No, actually, we are not,
because we are saying they have to demonstrate that
their change in performance is significant before we
go to them, and we're saying, what's the definition of
significant, it's consistent with the existing
philosophy, you've increased your change by a certain
DOCTOR KRESS: Yeah, but you could allow
those plants to degrade their performance without
worrying so much about it.
MR. MAYS: We're taking the same absolute
change in performance for all the plants.
DOCTOR KRESS: I understand.
CHAIRMAN APOSTOLAKIS: I think you made a
good point, Steve, but maybe we ought to think a
little more about Tom's point, too, but I think your
point is well taken.
DOCTOR KRESS: Yes, I think it's not a bad
point, I'm not totally disagreeing with you.
CHAIRMAN APOSTOLAKIS: But, here is the
place where I think we can revisit the question of
putting constraints on the proliferation of the number
of RBPIs. You state in the report that the loss of
feedwater and loss of heat sink are performance
indicators that are not in the existing Revised
Oversight Process, and they just talk about
MR. MAYS: Well, actually, they have two.
CHAIRMAN APOSTOLAKIS: Unplanned SCRAMs.
MR. MAYS: they have three in the
initiating event cornerstone, they have unplanned
SCRAMs, which is just a count of all the SCRAMs.
CHAIRMAN APOSTOLAKIS: Yeah.
MR. MAYS: They have the number of
CHAIRMAN APOSTOLAKIS: Significant power
MR. MAYS: power changes, and they have
one that kind of represents feedwater and heat sink
CHAIRMAN APOSTOLAKIS: Right.
MR. MAYS: So, this one is
CHAIRMAN APOSTOLAKIS: But, the question
is, I think this is where we could ask the question,
is it worth treating them separately, so that the
number of performance indicators increases to the
dismay of the industry?
MR. MAYS: Actually, in this case the
number wouldn't change.
CHAIRMAN APOSTOLAKIS: But, why?
MR. MAYS: If you had three, you would have
three, so there wouldn't be any net change if you were
to make a complete swap out.
CHAIRMAN APOSTOLAKIS: In terms of
collecting data, it wouldn't make any difference, I
MR. MAYS: No, and it wouldn't make any
CHAIRMAN APOSTOLAKIS: But, in terms of
having more indicators it really does make a
difference. You have three now, they had only two.
MR. MAYS: They had three.
MR. HAMZEHEE: They have unplanned SCRAMs,
loss of normal heat removal pump and reactor power
CHAIRMAN APOSTOLAKIS: Where would you put
the unplanned SCRAMs, in general transient?
MR. HAMZEHEE: Yes, usually.
MR. MAYS: We would substitute general
transients for unplanned SCRAMs.
CHAIRMAN APOSTOLAKIS: Significant changes
MR. MAYS: We wouldn't use those because we
can't make a relationship between risk in that.
CHAIRMAN APOSTOLAKIS: So, in this
particular case you would preserve the number.
MR. MAYS: Well, that would be a decision
for NRR to say whether they were going to preserve it
or not preserve it.
CHAIRMAN APOSTOLAKIS: No, I understand
that, no, but let's not avoid the thing. I want to
get into the question of whether loss of feedwater and
heat sink, how can we scrutinize them? Let's say that
the other numbers don't change, or they change, what
kind of criteria would we be using to decide that,
yes, loss of feedwater deserves to be a PI by itself
because it gives me this information that I don't have
otherwise, or it does not because it doesn't really
You know, this is, I think and we don't
necessarily have to have the answer today, but I think
it's an important question.
DOCTOR KRESS: But, I think the answer is,
is it by itself risk significant?
CHAIRMAN APOSTOLAKIS: Well, actually,
Hossein gave an answer, he said that their risk
implications are different.
MR. HAMZEHEE: And, that's the main reason
for this study to treat them separately.
CHAIRMAN APOSTOLAKIS: And, you should
emphasize that and tell the NRR folks that this is an
important consideration, that it's not just the number
of the PIs that matters.
MR. HAMZEHEE: Correct.
CHAIRMAN APOSTOLAKIS: But, in this
particular case you are also eliminating one or two,
but in others you might not, although I didn't see
again, Steve would say that that's for NRR to decide.
MR. LEITCH: But, are we losing some
significant piece of information by eliminating
unplanned power changes? Say it again, you can't draw
a connection between that and the risk?
MR. MAYS: I'm saying, I don't have to
go back to the three things I needed to be able to do
an RBPI, I have to have a model that reflects plant
risk, I have to have a baseline performance that
allows me to make changes to that model to set
thresholds, and then you have performance data. I
can't make, in my risk information, a link between
going from how often I go from 80 percent to 30
percent power at the plant to what that has to do with
risk. And so, therefore, I'm not able to make a risk-
based performance indicator from that.
But, whether or not that means that that
PI is useful for other reasons is something that NRR
would have to decide, as to whether or not they wanted
to keep it, or not keep it, and I'm not making that
judgment here. I'm saying what risk-based performance
indicators am I capable of putting into play.
MR. JOHNSON: Yeah, and that's I'm
sorry, Steve, I just was anxious to add to the point
that you were making. You know, if you look at some
of the PIs that we have now, we've said that they've
not been not all of them have been risk informed,
but, for example, we know that when you look back
historically at plants that have had a significant
number of power changes as a result of equipment
problems, to address those equipment problems, that's
indicative of a plant that's having problems. And so,
there may be a situation where you would have a PI,
even though from a risk-based PI perspective you
wouldn't have that PI, but because of what we are
trying to do with performance indicators, and
providing an indication of the old raw performance of
the plant, you might keep that performance indicator.
So, that's the kind of consideration that we'll go
through in the change process in deciding to what
extent we replace, or add, or whatever, check the PIs.
CHAIRMAN APOSTOLAKIS: But, this is where
we would like to see some more discussion of these
things, and limiting the number of PIs I think and
I think we already mentioned some very valid points.
So, out of curiosity, the number of
unplanned changes in power, significance changes in
power, what kind of an indication is that? I mean, if
it's not risk related, what is it then? Is it
MR. JOHNSON: Well, it sort of gives an
indication, I see Tom from NEI, Tom Houghton from NEI
raising his hand, I guess you've got to get near a
mic, Tom, to speak, we believe it gives an indication
of, yeah, things that are not steady state at a plant.
If a plant is having situations that require it to
undergo a number of transients, again, setting aside
those things that are not induced by the performance
of the plant, that are not being generated from some
outside influence, but if a plant cannot maintain
stable operations because they are continuously having
to respond to things that are unplanned, that's
indicative of a plant that's beginning to have some
difficulty and, perhaps, warrants some follow-up.
CHAIRMAN APOSTOLAKIS: So, it really has to
do with the culture.
MR. HAMZEHEE: Well, it's such an indirect
indicator, you just don't know what it's coming from.
You can't conclude that that kind of condition may be
indicative of some problem, it may be culture, it may
be whatever, but it's actually, you know, an indirect
MR. HOUGHTON: Tom Houghton, NEI. We've
found that it is an indicator. It is more predictive
of future problems, and it did have a good
relationship with plants which were on the watch list,
okay, when the historical data was looked at. Okay.
So, it has face validity, I'd say, and it is somewhat
predictive, in that if the operations or maintenance
are not able to maintain the plant at the power level
that was intended in the management plan for operating
the plant, that there is a necessity of looking into
the problem further. Now, some of the cases have
involved bio fouling in condensers that weren't being
looked at as closely as before, or feedwater control
problems that weren't being looked at as clearly as
before, and partly because they weren't part of the
design basis or the tech specs of the plant, and so
this has led the plants to make a closer look at how
they are operating and maintaining beyond what's
So, we see some value in that. Now, there
have been questions about why 20 percent, why 72
hours, et cetera, et cetera, and there are efforts in
piloting revisions to this indicator which NRR is
proposing to pilot and industry is looking at a
similar pilot to try and avoid some of these questions
that have arisen as to what was intent, because you
want to try and take what was the intent out of it.
But, we've found that it's been valuable in the
CHAIRMAN APOSTOLAKIS: I guess what you are
MR. HAMZEHEE: One thing I would like to
note, however, on that example, dependency on watch
list, I looked through the data, too, and often times
a plant has a lot of power changes after it gets into
the watch list, which means the operators are
sensitive to regulatory observations that suddenly,
truly, I mean, it is so transparent, you know. So,
that's why I'm saying it's such an indirect indicator
that, you know, it's very hard to fathom what is
causing it, and if it is clearly there are
implications, because if you have a lot of power
changes they may initiate a transient of some type.
CHAIRMAN APOSTOLAKIS: So, what I gather
from this is that we just found a performance
indicator for the crosscutting issues. Well, that's
what you told us. So, if the maintenance department
doesn't do a good job
MR. JOHNSON: George, we actually think
that the full spectrum of performance indicators and
the inspectible area results provide a good indication
of crosscutting issues, in that
CHAIRMAN APOSTOLAKIS: I didn't say it's
the soul indicator, but it's an indicator.
MR. JOHNSON: in that problems
CHAIRMAN APOSTOLAKIS: Why are we so afraid
of this safety-conscious work environment, every time
I mention it I get no, ten nos. Why? Is there
something magical about it?
MR. MAYS: I've never given you a no on
CHAIRMAN APOSTOLAKIS: Otherwise it would
have been 100 nos.
MR. MAYS: Right, I think you are seeing a
consistent situation here, George, and that is we
don't have anything that goes up and says this is our
direct indicator of safety-conscious work environment,
because we don't know how to measure it that way.
CHAIRMAN APOSTOLAKIS: I didn't say it was
MR. MAYS: But, what we have is
CHAIRMAN APOSTOLAKIS: I didn't say it was
the only one.
MR. MAYS: what we have is, multiple
ones, that's why we took the sample approach, and
that's why we are seeing that there are some cases
where we have things that help us in that area, and I
think that's appropriate. I don't think we should be
afraid to say, my personal opinion is, I don't think
we should be afraid to say that we have a sample of
things, and we have some that are more direct than
others, and giving us indication when that particular
area is having difficulty. I think we have those.
DOCTOR KRESS: I think you do have, and I
think the question of that bears on the question
of, is there an optimum number of PIs, and normally
when you ask that question, is there an optimum number
of PIs, when you relate to other statistical treatment
of things you are talking about a sample and how many
samples do I need to have the confidence level that
I'm measuring what I think I'm measuring.
And, in your case, I don't think you have
the capability of determining that optimum, and when
you can't determine an optimum in a statistical
manipulation or looking at the data, I think you have
to just fall back on take as many as you can. I hate
to say this, because the industry, you know, I can see
them now, but if you can't determine an optimum from
statistical analysis of the thing, then it seems to me
like the only other option you have. I'd be
interested in hearing your reaction to that.
MR. MAYS: Well, I think, I'm not sure your
taking as many as you can is necessarily the answer.
I think the problem is, you are trying to reach a
question of figuring out when you reach the point of
diminishing returns, and sometimes you can do that
because you have data and information on a model to do
that very precisely, and sometimes you have to do that
from a more judgmental approach.
DOCTOR KRESS: I count that in the phrase
as many as you can, I mean, that's part of the as you
MR. MAYS: I think the bottom line at the
end of the day is, do we have confidence that we have
a process by which we can detect when plant
performance is degrading from a safety standpoint, so
that the NRC can take appropriate action to intervene.
DOCTOR KRESS: How can you validate that?
MR. MAYS: Now, the question for that one
is, I don't know that you can do the kind of
statistical validation of that that might be desirable
to do if you could, but we have a philosophy that says
we want to try to have objective, measurable, risk-
informed information to do that, and I think, again,
this is part of that progress versus perfection
discussion, we will have more of it here, and then it
has to become a judgment as to whether or not we are
achieving much benefit when we do that. That's part
of what the ROP process has as their joyful task to
figure out, as part of the change process.
The next thing we wanted to show you was
the results of some of the work from the mitigating
systems. We had proposed in the RBPI report that we
could come up with 13 mitigating system component
class RBPIs for BWRs and 18 for PWRs. These were
using the SPAR models again for setting the baselines
and the thresholds. We used the system reliability
and component reliability studies that we've produced
in Research and formerly in AEOD as baseline
information to go into those SPAR models. We used the
EPIX data for calculating reliability parameters, and
we used the current information that's coming in
through the Reactor Oversight Process for putting the
unavailability data into these models.
So, the point here is, this EPIX data for
the reliability is the only part of this that is data
that isn't already reported to the NRC in some quality
fashion that we already know about, so this is the one
where we have the implementation issue.
And, we used a similar process that we did
for the initiating event indicators, for figuring out
what the time frame and the right prior was to do
that. And, when we did that, because we had and
when you get to reliability, if you look at
reliabilities of pumps, and diesels, and other things
which have generally mean reliabilities in the
vicinity of E-2 or potentially lower, we found that
even with a three-year type of time period we still
had situations where we would have false positive
rates that could potentially exceed the 20 percent
that we had set up as an initial basis. And, what we
decided to do with that, and you'll see it in the
tables as we flip back in a minute, is that whenever
we had a reliability indicator that crossed the
green/white threshold, we would also add an additional
piece of information which is, the probability that
the baseline value was still below the threshold. So,
basically, recognizing that the probability was a
distribution, sometimes the delta between the baseline
value and the green/white threshold was fairly small,
it would be easy for that distribution to cross the
threshold and we wanted to make sure we gave a little
more information to say, well, is it like really
across the threshold or is it just barely across, so
we gave a little more information because we couldn't
always meet the 20 percent false positive threshold
that we used.
DOCTOR KRESS: And, will that information
be used somehow in the overall plant assessment
MR. MAYS: Well, we thought that that was
appropriate to use because
DOCTOR KRESS: It's good information, I
MR. MAYS: we wanted to have some idea
of how sure we were that somebody had gone over that
DOCTOR KRESS: I think some guidance needs
to be developed on how we use that.
MR. MAYS: I think that would have to be
done on how to use it, but we thought it might be
something that we talk to people about.
DOCTOR KRESS: I definitely think it's
useful additional information.
MR. MAYS: I'm going to skip on over to
page 18 now and show you what we had for the
CHAIRMAN APOSTOLAKIS: Oh, in general,
reading from the original report, I get the impression
that based on the numbers you got you can actually
have threshold values for classes of components or
classes of plants, that you don't necessarily have to
have a separate threshold value for each component at
MR. MAYS: Well, we are looking we are
going to look at that.
CHAIRMAN APOSTOLAKIS: Is that correct?
MR. MAYS: Right now, we have only 23
plant-specific models for which we've done this, and
one of the things we said we would go back and look at
was, was the differences among the plants or within
groups so much that you had to do a plant-specific
value or whether it makes sense to make a group value.
We haven't got all that information to be able to do
that yet, but that's one of the things that was
proposed as a way to deal with the number of number
and types of PIs.
CHAIRMAN APOSTOLAKIS: Which is very
annoying here, I guess, but not the Maintenance Rule,
which is another mystery to me. Why in the Maintenance
Rule the licensee was asked to submit plant-specific
thresholds, and everybody thought it was great, but
when it came to the Revised Oversight Process it's
something that is like, you know, we don't want to
MR. MAYS: One of the comments we've
received from industry was a concern that if we have
risk-based performance indicators set up this way that
there might be a potential conflict between thresholds
here and values set for the Maintenance Rule.
CHAIRMAN APOSTOLAKIS: And, that's
something we cannot resolve?
MR. MAYS: No, we could potentially resolve
CHAIRMAN APOSTOLAKIS: Yeah.
MR. MAYS: The issue has to do, almost from
a technical standpoint, to do with the fact that in
this case we are doing a single-point variate
analysis, we take one thing, we hold everything else
constant, and we see what the impacts are.
Aside from the fact that they may be using
a slightly different risk model at the plant than we
were using, that's one of the bigger issues. One of
the things the plants were able to do, because they
were having a more integrated look at this, was to
say, for example, okay, suppose I desire to be able to
have a greater unavailability of my diesel generators,
because I have a financial or other reason for
conducting some on-line maintenance, well, what I will
do is, I will trade that off by making sure I have a
stricter standard for my reliability, so that in net
the risk hasn't changed significantly.
Well, if you have a single variable
analysis like we have here, you can't make that
tradeoff, because you are holding all the other things
constant, and what we will see when we get down a
little further in here, we talked about ways of
reducing this, one of the things we are proposing is
a more integrated way of looking at them, which can
allow for that kind of stuff to go on.
So, the Maintenance Rule was for the
licensees to set their own standards and for us to
monitor that they were doing those. So, I think
that's probably the answer why they didn't have a
problem at that level, because they were setting it on
their own standards, using their own risk information,
and being able to trade off back and forth where they
Anyway, coming to this example here, I
didn't want to go through all of the ones on each
case, I wanted to point out a couple things. One of
the things that you can see when you look at these
examples is that the case of the 95th percentile,
let's go down to emergency AC power unreliability,
what you'll end up finding there is a case, if you
take the 95th percentile, you get a value that's
almost up to the red value, if you were to take that
as your green/white threshold.
CHAIRMAN APOSTOLAKIS: Is that right? It's
close to where are you looking?
MR. MAYS: I'm looking at emergency AC
power unreliability, which is right here, this line.
CHAIRMAN APOSTOLAKIS: Oh, yeah, right,
because for the others that's not true, right?
MR. MAYS: Right, for the others it's not
necessarily true, so that was one of the reasons, an
example of a reason why we thought that the 95th
percentile approach may not be the most appropriate
way to do with these. So, that was an example.
CHAIRMAN APOSTOLAKIS: Well, I would say it
MR. MAYS: Another thing that we found when
we looked at some of this stuff is that sometimes,
because of the risk importance of a particular
component, even if its reliability or its availability
goes to one, it never producers the delta CDF
necessary to get to E-5 or E-4 to get you to the
yellow or red zones. So, that raises a question, is,
well, maybe we don't want to use that as an indicator,
or maybe we want to do something different.
We haven't come to a complete conclusion
on that, and sometimes what you'll see is, we'll find
that you can, in fact, reach those thresholds, but
only if you exceed te tech spec allowed outage times
for your equipment. So, the question is, do we want
to have an indicator that has a threshold that they
can only get to if they are violating the license.
I'm not sure that necessarily makes
CHAIRMAN APOSTOLAKIS: So, not reached
means not reachable.
MR. MAYS: Not reached has two things in
these tables. One of them has a footnote, I think,
which we eliminated the text on that, but the
footnote in the report, we have a not reached and we
have a not reached with a footnote, and we distinguish
between the ones that can't be reached because the
risk importance of the component isn't significant
enough, and those which it could be reached but it
would only be reached if you violated your tech specs,
in terms of operation. So, it's not really clear to
us which one makes the most sense here, we are just
laying out what the feasibility is of using an
indicator in that particular area, and what we were
trying to do is demonstrate that it's possible to set
thresholds for these particular values on a plant-
CHAIRMAN APOSTOLAKIS: So, what's the
difference between this and what the current process
has? Are you increasing the number of indicators?
MR. MAYS: Well, first off, we have
specific reliability indicators.
CHAIRMAN APOSTOLAKIS: That's correct.
MR. MAYS: We have availability indicators
and the reliability indicators have plant-specific
baselines and performance thresholds for them, and we
have, in another issue that we have, and we have a
broader coverage so we have more of them, and the
other thing we have, if you look at the bottom of that
page, we have three component class indicators for
air-operated valves, motor-operated valves and motor-
driven pumps, which go across systems. And, what we
have in that case is, we have a baseline value that we
get for the plant, and then what we have done is, we
said if we increased that value by a certain factor,
so, for example, the green/white threshold for AOVs
would be at 2.2 times increase in the baseline value
would get you to the green/white threshold, and what
we did there was, we took all the AOVs in the risk
assessment, said if we double them that's what gets us
to E-6. If we go up by a factor of 13, that's what
gets us to E-5.
CHAIRMAN APOSTOLAKIS: What's the point of
that? I mean, doesn't it go against what Hossein said
earlier, that not all AOVs have the same risk
MR. MAYS: Potentially, but what we are
trying to do here is say, if we had a broad
programmatic problem, if our AOV maintenance problem
was a problem, or general maintenance was a problem,
or we had a problem with our design and implementation
of motor-operated valves, if they were all to go have
a degradation, how much degradation would all of them
have to be going under in order to reach this
CHAIRMAN APOSTOLAKIS: So, that would be a
useful incite to the Option 2, no?
MR. MAYS: I don't know enough about that
to be able to
CHAIRMAN APOSTOLAKIS: That's a good
MR. MAYS: to say.
CHAIRMAN APOSTOLAKIS: Very few of us know
MR. MAYS: Okay.
Moving on to the next thing that we were
asked to look at by the user need letter, had to do
with containment performance, because there was a
limited number of things that we had in the ROP to
deal with containment. Unfortunately, we were able to
identify things that could potentially be used as
risk-based performance indicators for containment,
mainly the performance of the drywall sprays in the
Mark I BWRs, and the performance of large containment
isolation valves in the others.
DOCTOR KRESS: This information came out of
MR. MAYS: Right, these were the things
where it says what performance could have an
DOCTOR KRESS: Yeah, you don't deal with
those in SPAR.
MR. MAYS: well, not quite, when you
say SPAR, SPAR is a broad program, there is the level
1 SPAR models, there are LERF level 2 models, there
are shutdown models, and there are potential external
event stuff. So, SPAR represents that whole section.
DOCTOR KRESS: So, you are using the level
1 SPARs for this study.
MR. MAYS: We are using level 1 for the
initiating events and the mitigating systems, for the
containment we were looking to use LERF models.
DOCTOR KRESS: I see.
MR. MAYS: And, we are going to use LERF as
our metric, for containment related issues.
DOCTOR KRESS: And, that's another one of
my questions, but I'm sure you are going to discuss it
MR. MAYS: So, we were planning on doing
CHAIRMAN APOSTOLAKIS: Before we leave the
mitigating systems, there was a sentence in the
report, Appendix A, page A-25, "The same component
rate importance criteria were used to select class
indicators. However, the system level versus the
importance values were determined using the multi-
variable group function available in SAPHIRE." What
is this multi-variable group function available?
MR. MAYS: I think that's just a fancy way
of saying we changed all of the components to have the
same degradation at the same time, and in random model
again. Would that be correct, Steve?
CHAIRMAN APOSTOLAKIS: You have to come to
the microphone, please, and speak with sufficient
clarity and volume.
MR. MAYS: Fortunately, George, you and I
never have that problem.
MR. EIDE: Steve Eide from the INEEL, and
I believe Steve is correct. I don't know the
specifics of that actual
CHAIRMAN APOSTOLAKIS: Which Steve, this
Steve is correct?
MR. EIDE: Steve Mays, I don't the
specifics of that actual
CHAIRMAN APOSTOLAKIS: But, it sounds
MR. EIDE: module in SAPHIRE.
CHAIRMAN APOSTOLAKIS: Multi-variable
function, this is really impressive.
MR. MAYS: We have the capability in the
SAPHIRE code to go over and change multiple components
at one time with a change set, and that's what we
CHAIRMAN APOSTOLAKIS: Okay.
MR. MAYS: Moving to containment, the
problem we had, we were unable to develop containment
performance indicators, because we don't have the
models and the data currently available to be able to
do that on a broad enough either on a plant-
specific basis for sure, or on all the different
classes and types, so we were limited there by our
capability right now to be able to produce performance
indicators for containment.
DOCTOR KRESS: You do what you can, is that
MR. MAYS: That's correct.
DOCTOR KRESS: But, I would like to you
are not necessarily going to limit to LERF when you
get around to doing it. You mentioned that
MR. MAYS: Our original intention was to
use LERF as the metric for the containment
performance, because that would be consistent with
what we have in Reg Guide 1174 and other applications.
It's potential that we could go to something different
from LERF if somebody thought that that was useful and
worthwhile, but right now that's what we were looking
at on the basis of what we have available.
DOCTOR KRESS: I know a few people who
think that it would be worthwhile to include LERF
is all right, but consistency, you know, is the
hobgoblin or something or other.
MR. MAYS: Foolish consistency is the
hobgoblin of small minds.
DOCTOR KRESS: But, I think one ought to be
concerned in the regulatory arena with late
containment failures and also
MR. HAMZEHEE: In Phase 2 we are going to
look into this, to see if large late failures are also
DOCTOR KRESS: and I think you could
probably deal then with just the conditional
containment failure problem then.
MR. MAYS: Right. The issue then again
comes to, do we have a set of models that reasonably
reflect some understanding of the risk that we can put
data through and do, and right now we are just not
DOCTOR KRESS: Yeah, well, you know, my
I'm urging you not to think of risk just as prompt
fatalities, that's my point.
DOCTOR BONACA: Just one comment I have,
and probably I'm wrong, but because in many cases you
cannot really identify a meaningful RBPI, you simply
don't do that, and then you take the opportunity, you
know, to develop what you can get, but you want to
make sure that what you can get is meaningful, too,
right? I mean, what I'm trying to say is that, I'm
left with the impression that, you know, because of
that you are going to get a set of indicators that may
not be so significant after all, but the reason why
you got to those is because that's all you could get.
DOCTOR KRESS: I think one of their
criteria was, they have to be risk significant.
DOCTOR BONACA: They have to be, okay, but
I'm trying to understand the time, you know, how many
facets of this thing you are going to see, just maybe
two or three, and, you know, does that give you the
picture you want, or is it just all you can get. And,
I'm not sure they are the same thing.
CHAIRMAN APOSTOLAKIS: There is some of
that, this is a significant step forward, though.
DOCTOR BONACA: Oh, no, I'm not
CHAIRMAN APOSTOLAKIS: This could be never
DOCTOR BONACA: I understand that.
DOCTOR KRESS: Progress is what we
CHAIRMAN APOSTOLAKIS: Progress, we work
MR. LEITCH: Wouldn't performance on
integrated leak-rate tests be a significant PI in
MR. MAYS: I think that's been looked at
before. You could go back and look at performance
under leak-rate tests. The problem we've had in
looking at performance under leak-rate tests is, you
might be able to see that leak-rate test performance
has changed, but the question is, what's the risk
significance of that information? And, when you look
at the risk assessments and things that have been
done, the leak tightness of the containment in the
kinds of things that those things measure is rarely,
if I'm aware of, ever the dominant contributors to the
DOCTOR KRESS: It's never even risk
MR. MAYS: It's not even close.
DOCTOR KRESS: But, that's when your risk
measure is prompted out.
MR. MAYS: That's correct.
DOCTOR KRESS: So, that's why I'm saying,
don't just focus on prompt fatalities.
MR. MAYS: Right.
DOCTOR KRESS: You might want that as one
MR. MAYS: But, even in the cases of when
you look at latent cancer deaths and risk significance
DOCTOR KRESS: It's not significant there.
MR. MAYS: it's not significant there
DOCTOR KRESS: It's a risk of possible land
MR. MAYS: Maybe, but I'm saying, comparing
to the other things that would do land contamination
DOCTOR KRESS: That particular one is a low
MR. MAYS: it's pretty small, too.
DOCTOR KRESS: But, late containment
failure now is a different issue. It can be risk
significant from the standpoint of cancers and land
contamination. So, you know, but you are right on the
leak rate, unless it really gets bad.
MR. MAYS: The way it really gets bad is
somebody leave some major valve open, and that's what
I'm saying we would have
DOCTOR KRESS: You would capture that
MR. MAYS: Right.
MR. HAMZEHEE: And, that was one of the PIs
on the Reactor Oversight Process, but they also
eliminated that from the list.
MR. LEITCH: Okay, thanks, I understand.
CHAIRMAN APOSTOLAKIS: I wonder whether it
would make sense to take the set of the performance
indicators we have, or we will have, and go to a real
accident or incident, and see whether, like Three Mile
Island, whether you would see any change in these
things before the incident occurred.
MR. MAYS: I'm going to show you something
that directly relates to that in a little bit.
CHAIRMAN APOSTOLAKIS: Good.
What did you say, Tom?
DOCTOR KRESS: We probably don't have the
data for Three Mile Island.
CHAIRMAN APOSTOLAKIS: Well, for something,
something, I mean, to validate that this process would
MR. MAYS: Your point is, if there is
something that is dominant contributors to the risk,
are we having measures in our PIs that relate to
those, and I've got a particular slide that shows
CHAIRMAN APOSTOLAKIS: Okay, so I'll wait
until we come to that then. Okay.
Is this a good place to take another short
MR. MAYS: Sure.
CHAIRMAN APOSTOLAKIS: Okay. Then, we're
taking a seven-minute break.
(Whereupon, at 10:53 a.m., a recess until
CHAIRMAN APOSTOLAKIS: Okay.
MR. LEITCH: Can I ask just one question
for understand here before we get started again, or as
we get started again?
CHAIRMAN APOSTOLAKIS: Please, quiet.
MR. LEITCH: I'm looking at Appendix A, and
there's a number of pie charts
CHAIRMAN APOSTOLAKIS: Page?
MR. LEITCH: I guess it's actually
Appendix D, page 56, where the pie charts begin.
CHAIRMAN APOSTOLAKIS: Okay.
MR. LEITCH: And, I want to be sure I'm
correctly interpreting this information, just to pick
page 56 as an example, I think that's the first one,
it says areas not covered, 3 percent, indicators 2
percent, industry-wide trending 95 percent. Does that
mean, am I correctly interpreting that that 95 percent
of the issues are so infrequent that they are not
amenable to individual plant performance indicators,
that they have to be trended on an industry basis?
MR. MAYS: That's close.
MR. LEITCH: Okay.
MR. MAYS: When you look at the look at
what was in the IP database for the core damage
frequency associated with initiators for this
particular plant, what you find is that 95 percent of
the sequences involved an initiators other than the
ones we have direct indicators for, or the ones in
areas not covered. So, this might be a plant, for
example, that had really high contribution from loss
of off-site power events, or station blackout events,
since we don't have an indicator on a plant-specific
basis for that kind of thing, that would have to be
covered through the industry-wide trending. That's
how you would be able to see whether or not you
thought you had a problem, plus the plant-specific
inspections and the baseline inspections would go and
look at the areas that are not covered by indicators
to see if the performance that would impact were
changing. So, this is just to kind of give you you
are right, you are getting kind of the feel for which
portions of the initiating event indicators of the
risk are covered by the indicators, and which portion
would have to be either done by inspection and/or
DOCTOR KRESS: But, what is this a
MR. MAYS: Percentage of total CDF.
DOCTOR KRESS: Percentage of total CDF.
MR. MAYS: Right.
MR. LEITCH: Okay, thanks.
MR. MAYS: Okay.
Moving on to shutdown, this was an
important area because we didn't currently have in the
ROP any shutdown direct indicators. We looked at that
and we found that we couldn't do initiating event
indicators for shutdown because they just don't happen
frequently enough, but we did come up with some fairly
interesting things to do with respect to mitigation.
And, this has to do with several things.
We formulated a process by which we would
take into account the RCS conditions, vented, not
vented, open, not open, time after shutdown for decay
heat purposes, the availability of mitigating system
trains in those particular scenarios, and then we were
able to go back and try to set thresholds and
This one is a little different in picture
than what we had before, where we are actually going
out and calculating reliabilities and calculating
availabilities, now what we are doing is we are taking
a slightly different approach that says, if I have a
model that represents how a BWR or PWR responds can I
get groups of things, where if I spend time in those
scenarios I know those contribute more or less to
risk. So, we came up with for both the BWRs and
PWRs, were able to come up with thresholds. We put
together, started off actually with three categories,
low, medium and high, corresponding to an amount of
increase in CDF per day associated with being in those
CHAIRMAN APOSTOLAKIS: You know the
question you are going to get from some of my
colleagues in May, how can you do this if you don't
have a good shutdown PRA?
MR. MAYS: This goes back to the first
point I made earlier, in order to do the risk-based
performance indicators I have to have a reasonable
model of how a plant responds.
CHAIRMAN APOSTOLAKIS: Do you think you
have it now?
MR. MAYS: I think I have it for these two,
because what I have in these two cases is a plant-
specific model from a representative PWR and BWR, that
they happened to use for doing their shutdown for
shutdown risk models. So, I think I have something
that's reasonable here that I can use. I don't have
something for every plant. I don't have the SPAR
models developed for every plant, or even for the
groups of plants yet, but I have this information
that's a starting point for progress, not perfection.
So, that was the basis for doing this.
So, when we looked at that, we said, all
right, let's go back to the baseline and say, how much
time do these people typically spend in various
configurations, because it's just necessary to go
through some of them in order to complete a shutdown.
And so, we would measure performance as being
deviations from the nominal performance that you have
to do, just to go and conduct a shutdown operation,
and if you spend more time in particular
configurations of higher, lower, or medium risk
significance, then that would be the basis for us
deciding what the thresholds would be.
And, when we came to that, we also
recognized that for PWRs there's a special category of
the early reduced inventory situations that they have
to go into in order to be able to do that, which
represents a higher risk significance than most of the
other configurations they go to, and because when they
are in that particular mode they are under the
shutdown guidance of NEI guidance on what was the
number of that, Tom, I can't remember?
MR. HAMZEHEE: It's 91-06.
MR. MAYS: 91-06? 91-06, which says, when
you are going into early reduced inventory modes you
have to take certain compensatory measures with
respect to availability of power, availability of
injection systems, so we said if you are complying
with that in your early inventory, and you don't spend
more than the nominal time, then we will do that. If
you spend more than nominal time in that one, then we
treat that as if it's a high. So, we treated that
category a little differently.
DOCTOR KRESS: That's a really different
concept than what you did for the others.
MR. MAYS: That's correct, because this is
all we were able to do with the information we had.
DOCTOR KRESS: And, it goes back to my
concern about whether the baseline, which in this case
is called nominal, is sufficiently good enough, and
whether or not you are penalizing some plants you
know, if I were a plant that took long times at high-
risk significant configurations earlier, then I would
be able to continue doing that on this and not get a
white reading, because you are basing it on that as a
starting point. This part worried me more than any of
MR. MAYS: I understand your point. If we
had plant-specific history and plant-specific
DOCTOR KRESS: I understand.
MR. MAYS: values, that would be more
of a concern.
DOCTOR KRESS: Yeah.
MR. MAYS: I think what we are trying to do
here is say, what's typically representative of kinds
of times that industry generally spends in these
areas, so these baselines here were based on some
DOCTOR KRESS: They are basically industry
MR. MAYS: industry information, so if
somebody were to go and then start spending more time
in risk-significant configurations, they would not be
benefitted by having done that, the particular
arrangement you are talking about.
So, again, we are talking the progress,
not perfection, situation here. We have nothing now,
and we are trying to do something that's a little
better and a little more risk informative.
CHAIRMAN APOSTOLAKIS: So, the categories
low, medium and so on, are determined by the condition
of core damage probability?
MR. MAYS: Right.
CHAIRMAN APOSTOLAKIS: And, what are the
MR. MAYS: E-4, -5, or -6 per day, I
believe, that equate to core damage frequency for the
year of E-4, -5, or 06, if they were to spend their
time in that condition for a full day.
CHAIRMAN APOSTOLAKIS: Full day?
MR. MAYS: For a day. In other words, for
example, the high configuration would say, if you
stayed in that configuration for a day you would add
E-6, or E-4, to your core damage frequency associated
for that plant for that year.
CHAIRMAN APOSTOLAKIS: And, you would take
the day divided by 365 again, or that doesn't enter
MR. MAYS: We are saying that if you are in
a high configuration
CHAIRMAN APOSTOLAKIS: Yeah.
MR. MAYS: you are accumulating a
yearly increase of E-4 per day. That's the rate of
accumulation of the core damage frequency.
CHAIRMAN APOSTOLAKIS: Yes, I don't
understand that, but that's okay.
DOCTOR KRESS: It's like averaging it out
over the year.
MR. MAYS: That's right.
CHAIRMAN APOSTOLAKIS: Does the fraction of
one day over 365 enter anywhere?
MR. MAYS: Sure. What happens is, we base
all of our data gathering and our analysis on how many
days or hours you spend, and then the rate for the
high category is based on that translating to the
year. So, we do our calculations on the days, and the
rate for the threshold is based on the year.
CHAIRMAN APOSTOLAKIS: Right, and that's
wherein I offered the comment earlier I read, that no
matter how long you are there, if you divide by 365
you are effectively reducing its significance.
MR. MAYS: But, we weren't doing that.
That's what they didn't understand.
CHAIRMAN APOSTOLAKIS: Okay.
MR. BOYCE: It sounds like I don't have an
action item anymore.
CHAIRMAN APOSTOLAKIS: What?
MR. BOYCE: It sounds like it's understood
and we don't have an action item over on our side
MR. MAYS: I wouldn't be that bold.
The next thing I wanted to show was, let's
put up the PWR chart, we can go through this one a
little quicker because we have others to do. So,
basically, what we were talking about here is, you
would start on the left-hand side and you would,
basically, start at the top of the chart and move
yourself down and you would see for various different
configurations, like whether you are pressurized,
whether you are in mode 4, whether your reactor core
system was in tact, how many days after shutdown we
were, those are the going in conditions, for which we
then went and evaluated configurations and
combinations of configurations that previous PRAs on
shutdown have said to be important.
So, we would go back and, for example,
let's take an example for the one diesel generator, if
one diesel generator is out of service when you are in
pressurized mode for hot shutdown with the RCS in
tact, that constitutes the low category. So, what we
would do is, we'd gather up the amount of time you
spent in that low category, compare that to the
CHAIRMAN APOSTOLAKIS: Can you point to us
where you are?
MR. MAYS: Okay. I am on this row right
here, pressurized cooldown, Mode 4, RCS in tact, and
I'm looking at the impact of having one diesel
generator out of service.
So, we went back and looked at several
other configurations that were found to be important
to shutdown risk, relating to power availability, RHR
availability, secondary cooling trains, the
availability of the RWST, other things of that nature,
and we laid them out and if we have no entry in the
block then that means that particular condition does
not present a significant increase in the rate, so
what we do is, the nice thing about this, although it
looks busy, is that before the outage even starts,
when you've done your outage plan, you can go into
this table and see what equipment you are having out
when under what states, and before you even start have
an idea about where you could be accumulating more
risk than others. So, this is a nice tool for both
utility and the inspectors to have before you even go
CHAIRMAN APOSTOLAKIS: Isn't this similar
to what they are already doing with the various tables
they have risk configurations to avoid?
MR. MAYS: Correct.
CHAIRMAN APOSTOLAKIS: But, this is more
MR. MAYS: I'm not sure if it's more or
less detailed, but it has a similar concept. What we
are doing is saying, for a particular outage, we would
measure the time you spent in low conditions, the time
you spent in medium conditions, the time you spent in
high conditions, and we would compare those to the
thresholds, and that would tell us whether you were
spending excess time on those conditions, and then we
would know exactly what conditions we were in, we'd
know what to be able to go look for, so the idea here
is, is that you are going to be able to know in
advance what conditions to avoid. You are going to
know in advance what conditions you are planning to go
into, and the inspector, during an outage, if
something changes from the inspection plan, can go
right back to a table like this and say, all right,
now they are changing from this scenario to that
scenario, is that one I have to pay attention to and
worry about. And then, as we gather up the data as
you go through the outage, we can say at the end
whether your performance was basically nominal or
whether you accumulated enough risk in your off-
nominal conditions to warrant attention from the NRC.
That's the philosophy.
DOCTOR UHRIG: Would you get the same
information from one of these automated PRA
MR. HAMZEHEE: Not exactly.
MR. MAYS: Potentially, yes, I mean, but
I'm not sure exactly how they are gathering and
putting the information in, and how they are profiling
out the outage, but conceptually it's a similar
scenario. We are saying, what's the risk associated
with being in particular scenarios as we go.
DOCTOR UHRIG: Yes.
MR. MAYS: And so, it has a similar
foundation as the shutdown risk monitors in principle,
and so we think this is something that's fairly easy.
It's fairly easy to tell how much time you spent in
each of these configurations, because you planned it
all out before you start, and you monitor what you did
when you went through it, so if we were to get
information on how much time they spent in each of
these categories it would be fairly easy to do a PI.
Now, going back to the three things we
talked about earlier, having a model, having
performance data, we think we have, you know,
reasonable general stuff, and it is generic based more
than plant specific in this case. But, right now we
don't have data reported to us on the amount of time
spent in these things, so we would either have to have
somebody go out and get it ourselves, or we'd have to
have the industry produce it for us, in order to be
able to have a PI with respect to shutdown.
MR. HAMZEHEE: And, I think currently the
risk assessment that they do during the shutdown is,
they input all the equipment availability and how much
time they spent, and then they get a risk profile on
a daily basis, but they don't question as to how long
they should or should not stay in certain
MR. MAYS: I'm not sure whether they do
that or not.
MR. HAMZEHEE: But, they can do it if they
want to, they can go change some parameters and get
the results in the shutdown risk models.
DOCTOR KRESS: I'm particularly interested
in whether this will be part of the database you are
going to give back, because it's my view that you have
to have this information if you are going to do if
you are going to include shutdown risk within, say,
the 1.174 risk matrix, this doesn't do it by the way,
this information has to be fed back into some sort of
shutdown risk PRA in order to actually get the
contribution of shutdown risk to the total risk, and
also to determine the things like importance measures,
because this doesn't get reflected in importance
measures at all.
MR. MAYS: Indirectly it does. The
importance of the particular components, which is also
dependent on the particular condition that you are in,
is explicitly in the table.
DOCTOR KRESS: Oh, yeah, I'm sorry, it
doesn't show up in the importance measures you did for
the at power.
MR. MAYS: That's correct.
DOCTOR KRESS: Yes. I mean, you've got
some importance measures for components. This doesn't
reflect there in that.
MR. MAYS: Right.
DOCTOR KRESS: But, you do you are going
to have this kind of information for, you know, the
fleet of plants and for individual plants, if you are
really going to do a proper shutdown risk assessment.
So, I hope somebody starts developing a database on
MR. MAYS: Well, that's what we've
proposed, that if we have that kind of data we can at
least put together some information that would give us
indication of when something might be changing
DOCTOR KRESS: Uh-huh.
MR. MAYS: And, I think it's a good start.
DOCTOR KRESS: Yes.
MR. MAYS: Going on to the next thing,
which was fire events, this won't take very long at
all. Basically, the issue was from an initiating
event standpoint they don't happen often enough for us
to do plant-specific analysis of them. From the
mitigating system standpoint, we've identified what
systems would be important, which was the reliability
and availability of the fire suppression systems would
be the indicator we would try to put together, but we
really don't have the data to do that, to quantify
baseline and performance values, so
CHAIRMAN APOSTOLAKIS: Can we discuss a
little bit this issue of timely detection?
MR. MAYS: Right.
CHAIRMAN APOSTOLAKIS: And, I think it's
related to whether an indicator is leading or lagging,
is that correct?
MR. MAYS: No. Whether an indicator is
leading or lagging is what you are measuring and
comparing to. For example, all indicators, by
definition, which you gather from data are lagging the
data that you are getting, but they may be leading of
some higher order effect.
CHAIRMAN APOSTOLAKIS: Core damage
MR. MAYS: Correct.
CHAIRMAN APOSTOLAKIS: Core damage, yeah.
MR. MAYS: So, the issue here is, does the
occurrence rate of information for this particular
thing happen so infrequently, if I have, for example,
losses of off-site power which only happens in the
ball park of once every 30 years or so at a plant,
then I'm not going to accumulate data in a sufficient
period of time to be used effectively in the Reactor
Oversight Process, to go year by year and say to
myself, where does this plant need more or less
attention. So, I can't make that kind of an
assessment directly with an indicator for things that
have a low frequency of occurrence.
DOCTOR KRESS: Is loss of off-site power
under the control of the licensees?
MR. MAYS: Some of it is and some of it
DOCTOR KRESS: It could be a lack of, say,
a performance issue?
MR. MAYS: Right.
DOCTOR KRESS: Okay.
CHAIRMAN APOSTOLAKIS: But, is timely also
referring to the fact that if something happens it's
MR. MAYS: No, timely refers to the fact
that if there is changes in the performance, my sample
period is such that I can reflect and see that on an
ongoing basis and take action to deal with it on the
basis of that information.
So, if I have something like a LOCA
steam generator tube rupture frequency, or a LOCA
frequency on a plant-specific basis, there aren't
enough events coming along that allow me to trend how
that plant's performance is related to that particular
event. So, fire events comes in the same scenario
again, the frequency of fires at plants is low enough
that it's just not amenable to timely trending for
Now, we can do industry-wide trending on
that stuff, and we can cover the stuff that's not in
PIs through the inspection program, which is a little
more deterministic in some cases, the approach, but we
have a way to deal with them, but we don't have the
ability to do timely indicators of them, from an RBPI
DOCTOR KRESS: How would you get the data
in that middle bullet?
MR. MAYS: Data in?
DOCTOR KRESS: The fire suppression system.
MR. MAYS: Oh, if we were to be able to get
information from the plants on the number of times
that they find failures in the suppression systems, or
detection systems, the number of times they test them,
or demand them, those are the kinds of things the
same kinds of information we get for diesel
generators, or motor-operated valves, is not currently
DOCTOR KRESS: Do they test these fire
MR. MAYS: Some of them have testing
information, some of them don't. We have to see what
they have in order to see whether we can make timely
indicators. And, availability of these things is
something else that could be tracked, but right now
that information isn't being tracked and reported in
EPIX or any of the other stuff that we have
availability to, so we are unable to do indicators
directly on those.
CHAIRMAN APOSTOLAKIS: So, the response
time of the fire brigade during drills, that would be
an indicator if it were reported? Is it reported?
MR. MAYS: I don't recall that when we
looked at the fire risk assessments that the response
time of the fire brigade was a really significant
factor in the risk. I think what we found was, the
probability of detection and suppression was generally
more important, and I think Bob Youngblood has some
comments on that.
CHAIRMAN APOSTOLAKIS: Yes, but the
probability for suppression is really a judgment that
comes from the fact that you are going to have the
fire brigade, you are going to have CO2 systems and
all that. The problem is that the models are not
MR. YOUNGBLOOD: That's the point I was
going to mention, Bob Youngblood, ISL, we have
we've also had this report reviewed by fire PRA
people, and that's one of their comments. If we were
using IPEEEs in this, and they have a lot less detail
in that area, and one of our commenters said
specifically that he thought the fire brigade
performance was an interesting area, it's not, by the
way, equipment related necessarily, which would be
another desideratum, but he wasn't sure that the way
IPEEE has handled the whole thing that we were
necessarily getting the right perspective.
CHAIRMAN APOSTOLAKIS: Yes.
MR. MAYS: So, at any rate, we don't have
any fire initiators, fire initiator or mitigating
system PIs to propose to NRR, because we don't have
the feasibility to do them right now.
The next thing addresses, Tom, part of
what you just talked about in the pie charts. We
looked to see how much risk coverage the RBPIs were
giving us, and we took kind of two approaches, kind of
a Fussell-Vesely and a Risk Achievement Worth approach
to look at the thing. So, let me flip back to the
next table on page 27 here. What we went and said,
let's take a look at the information that's in the
SPAR models for the level 1 stuff that we were looking
at, how many events are actually in that model, and
relating to initiating events and cornerstones, and
how many of them are ones that we would be able to
cover using RBPIs, and you can see that we have a
percentage of those inputs into the total SPAR model
that would get covered by RBPIs.
But, the more important one, I think,
which addresses the question that George raised, is
the next chart, and that one is, this is one that I
think addresses the question that came out earlier,
and that is, we went back to the IP database and
pulled out the dominant sequences for each of the
plants that we were working on here, and looked at
what was the general things that were important to
those sequences. And, what we've done is, we've drawn
a box around all the pieces of the sequence for which
we either have an RBPI from initiating events or
mitigating systems, or we have an industry-wide trend
So, what you can see when you go down this
list is that most of the dominant sequences have one
or more pieces of them covered by an RBPI in this set
that we have looked at. So, that's a pretty warm
feeling to have, to know that you don't have a lot of
dominant sequences for which you've got no coverage at
all of your indicators.
CHAIRMAN APOSTOLAKIS: On the right, the
things you have boxed are part of the sequence, and on
the left the initiating events, what's going on there,
everything is boxed, but you have bold faced.
MR. MAYS: Well, we have two things. We
have bold ones were the ones we are directly having in
the RBPI indicators, the dotted lines ones under the
initiating event are the ones for which we don't have
plant-specific RBPIs, but we have industry trending.
CHAIRMAN APOSTOLAKIS: okay.
DOCTOR KRESS: I think industry trending is
a really good idea. I just don't see how it fits into
assessing the performance of an individual plant.
Will you touch on that after a while?
MR. MAYS: Well, this has to do with
something yes, we will, we've got some stuff on
industry trending in a minute, but the short answer to
that is, if I have to go and determine what's
important at a particular plant, and I don't have a
plant-specific indicator for it, then I have to ask
myself, well, what do I know additional about it, and
one of the things I might know is, well, over the
industry this particular thing, which might be risk
important, has been going up over the industry, maybe
that's something I want to pay more attention to.
DOCTOR KRESS: It just raises your
MR. MAYS: It raises your awareness.
CHAIRMAN APOSTOLAKIS: Increased monitor
MR. MAYS: And then also, if I have a
situation where I've seen a dramatic decrease in
something on an industry basis, then maybe I say to
myself, I don't need to spend as much time on my risk-
informed baseline inspection looking in those areas.
DOCTOR KRESS: But, what worries me there,
it's got compensatory errors, too, which bothers me,
some plants are going up and some are going down.
CHAIRMAN APOSTOLAKIS: See, that's his
MR. HAMZEHEE: But, we realize for this
event, though, Reactor Oversight Process, if it
happens once they are going to send a team to do a
root cause analysis, find out exactly what happened
and why it happened at a specific plant. So, this is
covered, but we don't have specific PI for it.
CHAIRMAN APOSTOLAKIS: But, you could also
do industry-wide trending for the stuff that you
MR. HAMZEHEE: And, we are going to, yes.
CHAIRMAN APOSTOLAKIS: And, that's useful
MR. MAYS: Right, we've got that in here,
The next thing we talked about is
CHAIRMAN APOSTOLAKIS: So, let's pick one
MR. MAYS: Okay. I'm trying to get you out
of here by 2:00, George.
CHAIRMAN APOSTOLAKIS: sequence nine.
MR. MAYS: Okay.
CHAIRMAN APOSTOLAKIS: All the way to the
right, it says "HUM," is that human?
MR. MAYS: Yes.
MR. HAMZEHEE: Yes.
CHAIRMAN APOSTOLAKIS: So, there is a human
action there, presumably, a dynamic thing, right?
MR. MAYS: Right.
CHAIRMAN APOSTOLAKIS: During to the
MR. MAYS: Right.
CHAIRMAN APOSTOLAKIS: And, there's nothing
we can do about it, right?
MR. MAYS: Well, that's not true. What we
are saying is that, the good thing about this table is
that these are the pieces of that sequence for which
I have direct performance indicators.
CHAIRMAN APOSTOLAKIS: So, the baseline
takes care of it, baseline inspection.
MR. MAYS: There you go, the baseline
inspection and any subsequent inspections should be
covering those areas for which I don't have a direct
CHAIRMAN APOSTOLAKIS: And, this is,
perhaps, NRR folks, this table, or a table like this,
it could be the basis for these tradeoffs that we
discussed earlier. If I put an extra performance
indicator somewhere, then I should reduce the
activities in the baseline inspection.
MR. MAYS: This is a similar concept which
CHAIRMAN APOSTOLAKIS: That's very useful,
MR. MAYS: right, this is a similar
concept that was used for devising the baseline
inspections in the first place. They went back and
looked at some PRAs, some stuff that was and wasn't
covered in the
CHAIRMAN APOSTOLAKIS: Not in such detail,
Steve, come on, not in such detail. I mean, it was a
MR. MAYS: It wasn't in that detail, but
the concept is the same, and what this does is provide
more detail they could be able to use as a basis for
going back and potentially looking at the inspection
MR. BOYCE: George, I think we agree
CHAIRMAN APOSTOLAKIS: Sure. No, I
MR. BOYCE: Well, the program has got to be
mature before we can really utilize the results with
any degree of confidence. We are not going to revise
our program based on preliminary results. I mean, we
are very interested in this sort of approach, and I
think in our initial comments, perhaps, even in that
aforementioned December 1st memo, we pointed out that
this was an area where we thought the program could be
And, right now, there's a separate program
outside of risk-based PIs to utilize risk incites in
our inspection program, and we've got that initiative
going in parallel to this, but it's not as systematic,
it's not as robust and detailed as this program has
the potential to offer.
CHAIRMAN APOSTOLAKIS: Good. Good.
MR. MAYS: The next thing we had in the
report was, we did some what we called validation and
verification, and what we wanted to do was go back and
prove that we could actually do this thing and produce
PIs and evaluate against thresholds, and so we went
back and used the 23 plants that we had for the period
1997 through '99, and put the data to the test to see
what happened. And, when we did that on the next
page, what we found out when we looked at that, we
think we have a more precise accounting for the risk
significant design features of the plants. We know we
have more plant-specific thresholds, and we think we
have a better dealing of false exposure time. That
was one of the things we mentioned earlier, and we
have this kind of "face validity" approach that we are
taking to say, does this stuff make sense from a risk
perspective, once we've put this stuff through the
models and seen what comes out.
So, we've got some tables to show you, and
we do have one caveat that we want to make sure we put
in here. We haven't had all this data and these
models go through peer review, so if anybody were to
conclude that this is a definitive statement that some
plant is either green or not green, that would be a
bad conclusion, because that's not something we are
trying to do at this point in time.
So, under the initiating events, we take
the 23 plants that we had, we've gone through and
determined what the actual data shows, we've got the
values in there, along with in the parentheses what
the particular color would be for those initiators.
On the next
CHAIRMAN APOSTOLAKIS: So, there are a few
MR. MAYS: Yes, there are.
DOCTOR KRESS: Is this a good argument that
George can use to say that the previous use of the
95th percentile was not an appropriate way to go?
MR. MAYS: Well, we've made that argument
generically in the report.
CHAIRMAN APOSTOLAKIS: They agree, I think.
MR. MAYS: And, that's the earlier tables
where we were showing the 95 and the other one was
based on more than that.
DOCTOR KRESS: Right.
CHAIRMAN APOSTOLAKIS: It's interesting,
though, if you look at I mean, there is a yellow
MR. MAYS: Yes, that's correct.
CHAIRMAN APOSTOLAKIS: That's B&W Plant 5.
MR. MAYS: Uh-huh.
CHAIRMAN APOSTOLAKIS: Yellow on the
general transient, white on the loss of heat sink, and
green on the loss of feedwater flow.
MR. MAYS: Right.
CHAIRMAN APOSTOLAKIS: So, what would the
Action Matrix dictate now?
MR. MAYS: Well, again
CHAIRMAN APOSTOLAKIS: That's beyond what
you are doing, right?
MR. MAYS: that's beyond what we are
doing now, and in addition, for that particular plant,
we were going back, remember I said we were doing
"face validity," we were going back and checking
because that looked to be higher than what we are
seeing on other B&W plants, and we're going back to
see if there wasn't a modeling issue that was causing
that to be, and we are looking at that as well. So,
that was the reason for that caveat in the previous
When we go to the mitigating system
unavailabilities, we have a similar layout for the
plants there on the next two charts. The key thing
there was that, for example, on AFW/RCIC, depending on
which ones you are PWR, we broke out the motor-driven
pump and either the diesel-driven or turbine-driven
pump separately, because that was one of the things we
found, that currently we were averaging trains
together, and they have different risk implications.
So, when we looked at them this way, we saw that the
risk implications were different, and that gave us
part of that face validity that we think we are having
something that makes sense from our understanding of
We also have tables for the unreliability
of the plants, and in this case, just to make the
table a little more presentable, instead of going out
and calculating all the individual mean values for the
unreliabilities for those sections, we know that if
you go over a three-year period we have no failures,
and any number of demands, that the update is going to
be equal to or less than the baseline, and since the
baseline was below green there was no point doing
anything more for it. So, we just took a shortcut in
this table and put less than baseline for all the ones
where we had no failures.
Now, if you look at that, for example,
down at the bottom, the PWR list, the last one,
Westinghouse 4-Loop Plant 23, you'll notice that in
the AFW column there is a number in there, and the
value is 1.5E-2 for motor-driven pumps, and then it
has an indication of white. And then you notice again
down below it there's a number, .13, that was the case
where we had something that went over white, and so we
said there's only a 13 percent chance, based on how
far that distribution overlapped that threshold that
the actual value was still at the baseline.
So, if you were to have, you know, a high
number like .87 or something like that, then you'd
say, well, maybe this isn't quite a white threshold,
maybe this is just the uncertainty in the data, but
when you have a fairly low number there you are more
confident that you've crossed the threshold.
I want to skip the last one, unless you
have a particular question on it. We had the
component class scenario, and we did a similar thing,
if we had no failures we didn't calculate the actual
number, and when we did have a failure in that group,
we calculated a number and determined whether or not
it was green, white or yellow.
The thing we've kind of touched on
tangentially several times here has to do with
industry trending. When we originally started out
this program, we were considering industry trending as
part of an integral part of the PIs, and then as we
looked at it more we recognized that it was related
but not directly a risk-based performance indicator,
or at least not on a plant-specific basis. But, we
thought it was important to capture that there might
be risk-important events that occur, and risk-
important activities that occur, for which we can't do
a plant-specific indicator, and it needs to be
captured someplace, it just can't be left alone.
So, for those we looked at doing industry-
wide trending, and what we proposed for industry-wide
trending, both as an input to the ROP for those areas
for which you don't have direct indicators, well, if
you don't know the specific plants is the industry
getting better or worse, that's an important thing,
and also because we have a requirement in our
strategic plan to report to Congress whether or not
we've had any statistically significant adverse trends
in industry performance, so we viewed this information
as also being an important piece potentially to that
performance measure for the Agency.
So, what we proposed in the next table is
that we would have and develop, and they are in
Appendix A, I believe, is where most of them are, you
would be able to trend all of the proposed RBPIs that
we've already got in the report, as well as several
indicators that had frequencies that were less likely
to occur, and we grouped them in this table by the
cornerstones that they impact, and what we would be
able to produce in each of those areas. So, that's
what we've proposed as potential industry trends, and
some of that information is in the report.
I think we've spent a significant amount
of time so far already, by the way, talking about
several of these issues, and again, these are
implementation issues that this report and this
program is not going to directly address, because we
are really looking at the feasibility of putting
together indicators for the ROP, but we recognized
these were important things, so in our interactions
with the ACRS, and the public, and other people, as we
were going along, we wanted to raise these issues and
get people's thoughts going on them so that we could
know what the perspectives were before we got too far
down the road.
So, the first question was, well, do we
even need anymore indicators at all, or are we okay
with the set we've got now? Can we do everything we
need to do and still get by? I think the answer to
that is pretty clear. We believe that we can run the
concurrent Reactor Oversight Process and do an
adequate job. The question is, can we do better?
And, the stakeholders had different views.
The industry said, well, if we are going to get
greater sample and more PIs then we there needs to be
changes to the inspection program as well. So, our
position is that these are consistent with the policy
statement and the concept to try to use more objective
risk information in all areas possible, and the ROP
change process is where we are going to make an
assessment of whether or not they are worth it. I
can't tell you all the details of how that is going to
come about, but, I mean, that's where that I can
tell you that's where the question goes to get
On the next one, the key issue was how
many PIs do you have? We have, potentially, you
could have, if you made swaps for like indicators and
new indicators, you could have in the ball park of
about 30 indicators per plant compared to the 18 that
the ROP currently has, and people are questioning,
geez, is that really an appropriate level?
Our position in Research is that the total
number of performance indicators should be
commensurate with the amount of risk coverage you want
to do by objective performance indicators, and that
number hasn't been determined, and that's kind of a
there is no magic formula for calculating that,
DOCTOR KRESS: It's a policy.
MR. MAYS: it's a policy kind of thing,
and that's something that once we see how much
coverage the RBPIs have, with respect to what the
current ROP has from the indicator standpoint, and
what the desired mix is between the two, somebody can
come to that decision, but we believe that's the right
DOCTOR KRESS: You just can't develop a
utility function for this, that's the problem.
MR. MAYS: Correct.
DOCTOR KRESS: And, that's what you need.
MR. MAYS: The next questions that come up
with implementation had to do with data sources, do we
have those data sources, do they have the required
quality in order to be used as part of the oversight
process? We believe that the key here is that the
data needs to be of sufficient quality so that if
there is an error in the data it's not going to change
your overall context of how you are going to view the
plant. So, for example, if somebody comes up and says,
well, I had reported 24.6 hours of unavailability in
my diesel and, in fact, you go back to the plant and
you found out that, wow, it was really 26, well, if 26
isn't sufficient to change your conclusion about the
plant we don't think that that's a level of precision
that needs to be part of the quality and the data to
do that. But again, this is another part of the
implementation issues that will have to get worked out
and we would expect that to probably get worked out
through a pilot program.
The next question that comes up has to do
with the next two, in fact, have to do with models.
I had pointed out earlier that one of the main things
was, you have to have a model that has a reasonable
representation of the risk. I chose that word
carefully, because we've developed level 1 Rev. 3 SPAR
models for about 30 of the plants now, and we've got
a program to develop them for the rest of them.
The number of models' needs depends on the
level of plant specificity that you want to have. We
may be able to group things, we may want to do them in
plant-specific, but that's something that we have to
eventually come to a decision by.
And, the external stakeholders recommended
that if we were going to use SPAR models in this kind
of a process that they be reviewed by the licensees.
We agree with that. We've already been in the process
of taking several of our SPAR models on on-site
visits, and we have got plans to try to do that for
all the rest of them, because we believe it's
important to make sure that we are not, you know, out
in left field compared to what the plants have.
Now, we've done ten direct on-site visits
to review SPAR models, and on some occasions we found
that there were either equipment or procedures to deal
or mitigate with certain sequences that we didn't know
about, and then once we found out about them we
included them, and sometimes we've been to plants
where we've gone and they've said, holy cow, we think
our model needs to be fixed, your's is a better
representation of what's going on here. So, there's
a difference in the way that things are done,
depending on how long it has been since the plant has
updated their IPE, and what their groups are, but the
key for us is that with the SPAR models we have a set
of models in which we have a consistent methodology
for examining the same kind of information across
CHAIRMAN APOSTOLAKIS: And, this is not
just for this project, I mean, this will
MR. MAYS: NO, this will apply to other
things in the agency, and I think that's an important
thing, also from a public confidence standpoint, that
I think we need to be able to say that we have
something that we look at that's independent of what
the licensees come and give us, so that we have the
ability to say we've done a critical look at what
they've presented to us.
The other advantage to us is that if we
have these models done this way, then we have the
ability when we have differences between their models
and our's to focus very quickly on what the basis for
the differences are rather than having to take a long
time to go over and review their model from complete
beginning to end.
CHAIRMAN APOSTOLAKIS: So, have the
licensees urged you to, in fact, have as the SPAR more
than their better IDs or PRAs? Some of the licensees
did a complete PRA.
MR. MAYS: That's correct.
CHAIRMAN APOSTOLAKIS: Have you seen any
desire on their part to have a SPAR model that you
have be the PRA?
MR. MAYS: Well, actually, what we found is
that, if we have significant differences between what
we have in our SPAR model when we go to a site,
between what they have, for example, in the core
damage frequency in our's, we sit down and say, what's
the differences, and if we find something in there
that is, well, we've put in this new system, or we
have installed these new procedures, or we've changed
the plant design from what you had here, we go back
and look at those things, gather that information, and
we make modifications to the SPAR models in light of
CHAIRMAN APOSTOLAKIS: But, I understand
the SPAR models are sort of approximate, or is that a
MR. MAYS: I don't think approximate is the
right word to use.
CHAIRMAN APOSTOLAKIS: Can I put a complete
PRA, like full scope, the Seabrook PRA, can I put it
in a SPAR model?
MR. MAYS: Okay. The SPAR model stands for
Standardized Plant Analysis Risk, that's our
determination of the style and method of doing PRA
analysis and we apply it across all the plants.
However, the SAPHIRE suite, which is the
engine that allows you to run that, is capable of
taking a plant-specific PRA and putting it into it so
that you could do that.
Now, again, the problem there is, and we
have several plant-specific PRAs that are available,
Research has put those available in SAPHIRE, the
problem again there becomes, that just represents our
version, that gives us a model that represents their
PRA, and that one from the next one to the next one
may have different HRA assumptions, different CCF
assumptions, different modeling assumptions that they
put into the plant. So, while we have the actual
model in that case, we don't have a consistent basis
across them for examining what's happening. So, I
think SPAR models provide a different kind of benefit
to us, because we know that if we go and look at
Westinghouse 4-Loop Plant A and Westinghouse 4-Loop
Plant B, that if there's differences in the CDF
associated with those SPAR models it's because we've
determined something different about the plants, not
because we have different modeling assumptions.
So, we tried to limit the impact of
CHAIRMAN APOSTOLAKIS: And, the
Significance Determination Process will be based on
the SPAR models at some point?
MR. MAYS: The Significance Determination
Process that currently exists is based on the SPAR
CHAIRMAN APOSTOLAKIS: It is?
MR. MAYS: It's based on the ASP and the
SPAR models, that's how it was developed, and the
Significance Determination Process for Phase 3, where
we go out and do a more detailed risk analysis out of
it than what's in Phase 2, which is the table lookups,
in Phase 3 we actually go and put together a model to
look at that, and in most cases that uses the most
recent updated SPAR models we have.
CHAIRMAN APOSTOLAKIS: So, the tables that
are being used in the SDP are based on the SPAR?
MR. MAYS: Absolutely.
CHAIRMAN APOSTOLAKIS: All right, let's go
MR. MAYS: A similar question relates to
the LERF models. We only have a limited number of
those available, and we only have a limited capability
to develop those in the short term, so the issue with
the LERF has to do with the RBPIs that there may be
some mitigating system components whose threshold is
set based on CDF, that when you consider LERF might
actually get different thresholds. So, we haven't
been able to do that yet, but we recognize that that's
an issue with respect to whether or not what these
represent the public risk with respect to the
So, let me get to the stuff which I really
wanted to talk to you about today, which was
DOCTOR UHRIG: After lunch?
MR. MAYS: Maybe after lunch, if you want
to talk too.
CHAIRMAN APOSTOLAKIS: Maybe we should do
that after lunch.
MR. MAYS: Okay.
CHAIRMAN APOSTOLAKIS: We can't finish
everything before lunch.
MR. MAYS: No, so let me leave with you
with a taste of what it is. What we've done here
CHAIRMAN APOSTOLAKIS: I want to go eat.
DOCTOR UHRIG: Let's have a taste of lunch.
MR. MAYS: You want a taste of lunch?
Okay, no problem.
CHAIRMAN APOSTOLAKIS: Okay, we'll be back
(Whereupon, the above-entitled matter was
recessed at 11:49 a.m., to resume at 1:00 p.m., this
. A-F-T-E-R-N-O-O-N S-E-S-S-I-O-N
CHAIRMAN APOSTOLAKIS: Back in session,
continuing with Mr. Mays and Mr. Hossein Hamzehee.
MR. HAMZEHEE: Yes, sir, correct.
MR. MAYS: Okay.
The thing we wanted to talk about next was
some alternate approaches we've looked at in light of
the comments that we had about the number of PIs being
excessive or too many, and so what we went to do was
relooked at what was the basis for doing these in the
first place, and what we did originally was we
devolved risks into smaller pieces, and we set all of
our thresholds for risk-based PIs that are in the port
at the level at which the data was being collected.
So, if I had data on reliability, I had a threshold on
reliability. If I had data on availability, I had a
threshold on unavailability. And then, we looked at
how much of an impact that changes in those values
would have on accident sequence frequencies when we
We took a slightly different approach,
which I'm going to go through in these next figures
and talk to you about, so let me just put the figures
up and go through them. What we did was, we took the
accident sequences and we devolved them down into risk
areas at a more functional level, rather than at
reliability and availability of components, and then
we looked to find out what data we could do within a
particular functional group and then reassessed that
against our criteria for whether or not it was a good
So, if you start from that's the wrong
title, it should be Industry Risks instead of
Individual, I'm sorry. But anyway, industry risk
comes from all the plants together and individual
plant risk comes from containment, core damage and
health effects things, and so underneath core damage,
which was where we were primarily looking, we looked
at what were the big pieces under initiators and
mitigating systems that might be amenable to a
slightly different approach, which would reduce the
number of PIs.
So, under initiators we said, well, we
might be able to group those into three groups,
transients, LOCAs and special initiators, and we list
some of the values, some of the types of initiators
that might go under that, for example, under LOCAs you
could have small, medium and large LOCAs, you could
have steam generator tube ruptures, you could
potentially have other ones like very small breaks or
inter-system LOCAs, things of that nature. And, under
mitigation we took the approach which is on the next
slide, which was we put together just kind of a very
basic functional event tree that's generally
applicable for anybody, for example, in a PWR where
you have an initiating event, your first issue is do
you establish reactivity control, then do you have
secondary heat removal. If you don't have that, do
you have feed and bleed, and then you have
recirculation, cooling. So, at that functional level
we were trying to see what we could do to do RBPIs.
And so, our concept is that what we would
do is we would develop functional impact models at one
of those levels and we would take the inputs for
reliability, availability and frequency that currently
apply to that functional level and use those as feeds
in together. Now, this is a case where we are having
a multi variate sensitivity study instead of single
variable sensitivity study. So, at the level, say, of
secondary heat removal, we take all the things that
impact secondary heat removal, put them into that
model, and see what that would change to the baseline
core damage frequencies that way.
So, when we did that, we came up with
three different kind of levels at which we could
potentially put indicators together. We could put
together an indicator, for example, at the cornerstone
level. So, if you went to the initiating event
cornerstone we could have an indicator that said this
is the impact of all the different inputs at this
cornerstone level together. We could do that also for
the mitigating systems, or we could go to the
functional level and have somewhere between three and
five indicators at a kind of higher order value, such
as heat removal, feed and bleed, those levels, or we
could go back down to the component and train level,
which is where we currently have stuff in the RBPI
So, in looking at that, the way it would
look would be something like this. At the cornerstone
level, you would have, basically, two indicators. You
would have an indicator for initiating events, where
you would take the data associated with loss of
feedwater, loss of heat sink and general transients,
and you put them all together and run them through the
model and see what the output results was.
Now, what's different about this is that
in these cases, in all these functional cases, you
have the threshold being set at the output condition,
not on the input condition. So, currently in the
RBPIs we have a threshold for loss of feedwater, we
have a threshold for loss of heat sink, we have a
threshold for general transients, what we would do now
is take that data for all those data and say, what
would be the impact on the sequences for all of them
collectively. So, the threshold is now set on the
collective sequences, not any individual input.
So, you might, for example, have gotten
better on feedwater, or worse on heat sink, and
somewhere in between on transient, and you may or may
not get better or worse, depending on how that would
go. So, this is more like the integrated indicator
that we had talked about doing in Phase 2, but it's
not the complete total plant model version.
At the functional level, down from the
cornerstone level, we came up with two ways of
potentially doing this, and one was to take the
mitigating systems and group them by what initiator
they respond to. So, we would say, we'd take all that
data that we previously had in those 18 or 13 RBPIs
and we'd say, all right, which of that data when it
changes, how does that affect the transient sequences,
how does that affect the loop sequences, how does that
affect the LOCA sequences, and we just put them
through the entire model for all those things and see
what the impact would be.
DOCTOR KRESS: When you say how it affects
the sequence, do you mean how does it change the
sequence contribution to the CDF.
MR. MAYS: Right, collectively, together.
DOCTOR KRESS: Collectively, together.
MR. MAYS: Right. We take all, so in other
words we would take all the failure to start
information, all the failure to run information, all
the unavailability information for components that
affect loss of off-sight power sequences
DOCTOR KRESS: So, your threshold would be
a delta CDF.
MR. MAYS: a delta CDF for that
DOCTOR KRESS: Uh-huh.
MR. MAYS: Or, that group of initiators.
So, that way we'd say, okay, the
mitigating system performance for transients is this,
it's green, white, yellow or red. The mitigating
system performance for loss of off-site power is this,
even though they'd be using some of the same data they
have different potential risk impacts. That's one way
to look at it. We'll show you some results of that in
The other way to look at it, which is a
little more like the current ROP, a little more like
the SSPIs and other stuff that INPO has, is to group
them by their function, their system functions. So,
for BWRs, for example, we would say we have RCIC and
HPCI systems that kind of do high pressure
performance. We have diesel generators. We have RHR
systems, we have what we've referred to as
crosscutting, which is those AOVs, MOVs and MDPs that
go across all the systems at the plant, we say we
could take that group and run them through the model
for all initiators, essentially, and see what the
combined impact of those is on the output.
So, we did that. We did a trial on that
to look and see what it looked like. So, let me show
you the first one we have, which is what would happen
if you did this stuff at the cornerstone level. We
took a BWR and a PWR plant that we previously have in
the report and we ran it through and said all right,
if we put, for the cornerstone level for the systems,
what we have here is that you take this particular
plant, and you take its data on diesels, on HPCI, RCIC
and RHR, all those together, and take all those
systems, all those inputs together and see how that
mitigation comes out, this particular plant comes out
to be white.
For the initiators, which is the next one
down, the initiator impact says it is green, so from
that particular plant we could come up with a white
for the mitigating systems and green for the
initiating event cornerstone, at that level.
And, we have a similar thing we've done
for Plant No. 23. Now, we didn't actually know these
were going to come out white and green, they could
have come out both green, or both white, or something
else, it just happened to be done that way. So, this
says I could actually come up with a value of what the
performance was at the cornerstone level, if I wanted
to do that.
Now, we'll talk in a minute about what the
advantages and disadvantages of doing that are, but
that's what we could have done at that level.
The next one I have to show you is if we
were to take the performance of the mitigating systems
and group their impacts by the initiating events for
which they are supposed to function. So, the first
case says, for the BWR plant, this says the front-line
systems, which is the RCIC, HPCI and RHR, as well as
the crosscutting component group, for LOCAs, for all
the LOCAs that would be applicable to that unit, the
performance for mitigating LOCAs is green. The
performance for mitigating losses of off-site power or
station blackouts is white, and performance for
mitigating transients is green. So, this gives you a
little more information than what you got a minute
ago. At the cornerstone level, you just knew
something was white, but you didn't know what. This
one gives you a little more detail. It says the thing
that's important at this plant is that this
combination of performance for all these systems is
most important in loss of off-site power sequences.
DOCTOR KRESS: This means that you take all
of your you have to take all of your input data on
liability and unavailability and run it through
MR. MAYS: Run it through the model.
DOCTOR KRESS: the model at that point.
MR. MAYS: Right.
So now, this is different from what we had
DOCTOR KRESS: You ran the model.
MR. MAYS: Now, I'm using the model to run
the entire thing through to get the impact.
DOCTOR KRESS: To get the impact.
MR. MAYS: Because I can't do it correctly
I mean, the advantage to the other ones
DOCTOR KRESS: This takes care of my
MR. MAYS: Right, but it creates another
DOCTOR KRESS: Yes.
MR. MAYS: And, the problem it creates is,
before what we had was, we would set the thresholds by
using the model and then we'd just collect data and
compare the data to the thresholds, we didn't have to
go back through the model again.
DOCTOR KRESS: Exactly right, now you have
to go through the model every time.
MR. MAYS: Now I have to go back through
the model every time to do this, so that's a
difference, because I can't take into effect the
combined effects without putting it through the model.
DOCTOR KRESS: Yes, that's right.
MR. MAYS: So, this is more like the
DOCTOR KRESS: It's almost like the
MR. MAYS: You are right.
So, that's how you could do it if you
wanted to group the mitigating systems in accordance
with, for example, the initiator they respond to.
Another way to do it, which we have on the
next slide, is to do it by kind of the high-level
functions that the systems perform. So, for example,
again, the same plants, same two plants, what I see
now is that the electric power system for the BWR
plant, which would be these reliability and
unavailability combined now, is green, the HPCI, which
is the reliability and availability combined, would be
white, the RCIC is green, the RHR is green, and the
component groups is green. So, now I have a different
kind of perspective about the performance.
DOCTOR KRESS: But, none of this changes
the amount of reporting requirements.
MR. MAYS: RIGHT. This is with the stuff
that we already have, for the existing level we were
using for the RBPIs in the report, so we are using the
exact same data that we had in the report to do the
indicator a little differently.
So, let's, you know, having done that, and
you can see now you see that the thing that's causing
the station blackout loop sequences in the BWR plant
here to be high are the ones associated with the HPCI
system, not with the emergency power system, so
there's kind of advantages to going both ways if you
want to do that. So, we looked at what the potential
benefits of these things would be, and at the
cornerstone level the biggest benefit is, I've got a
single indicator. It's just, did this plant's
initiating event information and mitigation systems
that I'm monitoring rise to the level of having
performance that I'm concerned about. One indicator,
one time, see it. That's not too bad, and the other
advantage of this is that it takes into account inter
and intra-system impacts of changing in performance in
different areas, and we actually went back and looked
at this and we found, as we looked at the plants,
sometimes you would have things that were, say, in an
individual indicator, it was green and white, you put
them together in a combined thing and they turn out
green, or sometimes we'd find it went the other way,
there would be a white and a white, and you'd put them
together and it turns out yellow, or you find things
that are green and green and they turn out white, or
two whites end of turning I mean, you see variation
depending on which sequences particular inputs are
DOCTOR KRESS: Now the question I might
have is, what's the down side of doing all of these?
MR. MAYS: Of putting it all together?
DOCTOR KRESS: Doing all of them.
MR. MAYS: That's another that's another
thing you could potentially do. Let me get to that in
DOCTOR KRESS: Okay.
MR. MAYS: The limitation, of course, is
that once you find something that's not green
performance, you don't really know directly what it is
that's causing it so that you can go out and find out
what you need to spend regulatory attention on. It's
not very precise in telling you what's the particular
area that needs to be dealt with.
If you go to the functional level, well,
the benefits are we have fewer indicators, instead of
18 or 13 we are now talking about three, four, five or
six, depending on how you want to use these. This
also accounts for intra and inter-system impacts, and
it can be grouped either by major types of initiators
or by system functions, or if you wanted some other
way of looking at it you could postulate one and we
could do that.
The limitations of this is, now when I
have them broken down into functional groups, I now
have to have some way of bringing them back together
to make my assessment of whether the cornerstone was
degraded or not, because they don't directly tell me
the entire cornerstone picture. And, I still have the
situation where if I have greater than green
performance I still have to do more work to figure out
why it was greater than green. I may know it's in the
HPCI system, but I don't know if it's the availability
or the reliability, I've got to go back and do some
more looking before I can figure that out.
DOCTOR KRESS: That's why I was asking why
not do all of them?
MR. MAYS: Well, you are getting to the
The last one I had was, if you do it at
the component or train level, like the current RBPIs,
the biggest advantage here is this is the broadest
evaluation of individual attributes, and the causes of
greater than green performance are pretty obvious once
you've got it at that level. I know it's diesel
generator reliability, or I know it's AFW diesel-
driven pump train reliability or its availability. I
know the area much more precisely when I have the most
broad individual indicators, and it's much more
similar to the current indicators because the
indicator data and the thresholds are set and I don't
have to do running through models anymore, I just pick
up the new data and compare it, away I go.
The limitations are, the inter and intra-
system dependencies aren't accounted for here. So,
sometimes you get worse and sometimes you get better,
depending on what the risk relationships are on the
accident sequences, and if you have them at an
individual variate level you don't see that.
Now, the disadvantage also is it nearly
doubles the number of current PIs that we have, and it
requires you to set an individual plant-specific
threshold on lots of different indicators.
DOCTOR KRESS: But, it doesn't double the
quantity of data that you need collected.
MR. MAYS: It's the same amount of data.
DOCTOR KRESS: Same amount of data.
MR. MAYS: Exact same data.
So, that's the kind of stuff that we've
looked at as potentials, so we are looking for
Now, one of the things that you mentioned
that you could do is, you could say, well, if you are
going to take all that data and run it through the
model, either at some lower level, intermediate level,
or up to the cornerstone level, well, why not do it at
all of them and just have one of them be the one you
report out to the public, and utilities and everybody
else, and the other ones be the ones that you use as
subsidiary things to go back and see what was causing
it to be the way it is. That's a possibility. We
haven't really done much more than have some
preliminary discussions with NRR, because we just got
finished running some of these examples, as to what
one they think would be the best.
So, what we intend to do is try to get
your feedback on where you think we should go. We are
going to talk about this at the public meeting next
week, to see if people think that this is a good idea
that they would prefer or not, then once we get your
comments and their comments we are going to sit down
with NRR and we are going to say to ourselves
collectively, what makes the most sense to do and
publish in the Phase 1 report, which should be out in
November. So, we will have a kind of meeting of the
minds at that point, and we'll say, based on the
comments we heard, and our own internal discussion,
this is the way we think we should go. So, this
report could be dramatically changed if we decide to
go at a different level. If it's decided to throw
away the component level then this report would
certainly change, or if it's decided to do it at multi
levels this report will change, but we have to come to
that decision after we get some comment and feedback.
I think what I wanted to make sure you
understood today was we have the ability to do this in
different ways, and we are looking for feedback as to
what people think would be the best way to go.
And, we'll take those comments and we'll
go from there.
So, what we are looking for the ACRS to
give us feedback on, again, is whether they think
these represent potential benefits to the ROP or not,
whether they think this technically is an adequate job
of how you would go about defining and calculating
these things, and whether or not the alternate
approaches we just discussed here are something that's
worth pursuing or not, or whether it's something to be
not done until Phase 2, or done as part of Phase 1, or
where you think that kind of stuff should go.
So, I'm pretty confident that we have
information that uses readily off-the-shelf PRA tools
and methods, that uses readily available and
accessible data to us, that we can put together
broader and more plant specific performance indication
for potential use in the ROP. And, at that point if
we've got that, we'll hand this off to our friends
over at NRR, who can then go through the
implementation process, and where we will have to
answer the questions like can we actually get this in
the plants, what's it really going to cost to get this
data, are we willing to use these models, do we want
to instead use the plant-specific models from the
licensees by giving them some specification and using
those? I mean, those are all possible questions that
could get answered through the implementation, and I
don't want to minimize the fact that those are serious
questions and will require some serious work to make
it happen, but I think as long as we keep in the mind
set of progress, not perfection, and is this a valid
incremental improvement or potential improvement, then
that's where we want to be at the end of the day.
So, I guess the only other thing we have
is, if there's something I guess we have to hear
CHAIRMAN APOSTOLAKIS: Let's hear from Tom,
MR. MAYS: About what you want us to
present to the full committee.
CHAIRMAN APOSTOLAKIS: we'll do that
after we hear from Tom.
MR. MAYS: Okay.
CHAIRMAN APOSTOLAKIS: So, Nuclear Energy
Institute, please go.
MR. HOUGHTON: The ball moved down the
field, so I decided to take notes and talk from them.
The first thing is, the Nuclear Energy
Institute and the industry is very interested in risk-
based performance indicators and trying to move
forward in having the process be more risk-based as we
can. Of course, the caveat always and I think
Steve has done a real good job in working through this
and raising a lot of the issues in the analysis he's
done, and a number of things he's done address
problems that are problems with the current program,
which I want to talk to you about.
But, the caveat, of course, is that it
needs to be considered in the context of the ROP, and
what the purpose of the ROP is, and the performance
indicators. The performance indicators are meant to
be used to help the NRC determine how much inspection
it's going to do above the baseline inspection, and
how it assesses plants and how it engages in
enforcement of plants.
Our feeling is, if there's no reduction or
efficiency improvement in inspection, that it's
difficult to understand why we would put additional
effort into generating performance indicators.
Now, one could say that a lot of this
information is gathered under the Maintenance Rule and
under internal performance indicator gathering, and it
is. The problem comes, is that we move from, gee, I
think that was a loss of feedwater initiating event
to, my inspector is coming in and he's looking in a
manual and he's making a decision for which I can be
cited for a violation of the regulations in my
reporting, or it involves a long-winded process of
trying to resolve weather the issue counts or not, or
whether the hours count or not.
We've had on the order of about 260
frequently asked questions over the course of the
program so far, and these involve sometimes matters of
15 minutes of unavailability. So, the devil is in
these details, and as we expand the number of PIs it's
not just we might have some of that data, the question
is, is it worthwhile the extra effort that has to go
into that. That was one point.
We support a stable, consistent and
improving system of performance indicators, which Mike
Johnson talked about and Steve talked about, in terms
of the change process. I suspect that some of your
questions that I heard today are, why aren't we
thinking about some of these implementation or use
issues at this stage, rather than finishing up the
whole Phase 1 effort and then turning it over to NRR.
Unfortunately, to determine that this indicator
doesn't provide us additional value or a P process, or
that it's too difficult to explain to the public what
you are working about, because you are talking a more
sophisticated type of indicator, particularly, as I
was hearing if we went to a cornerstone level
indicator I think that would be more difficult to
understand, as opposed to a shutdown or an
So, we would and we support piloting
these. There are a number of pilots going on now.
There's one going on about the SCRAMs and the loss of
normal heat removal. There's one that's going to go
on to try and revise problems with the power changes
that will be piloted soon. And, the results, the
purpose of these pilots, as you were asking, is,
really, is it easy to understand what the indicator
is, is it easy to report it without making errors, and
is it more efficient in terms of inspection, is it
more focused on risk-significant issues, those are the
sorts of things that we want to see coming out of a
performance indicator so that it's of value to both
the NRC and to the industry.
A couple of comments on the initiating
event PIs. I think plant-specific goals are a good
objective, however, when you look at the range of data
it may become difficult to explain to the public or to
understand as a licensee the fairness of one plant
getting extra inspection if they had two general
transients, two SCRAMs, and another had seven. Okay,
that just you know, a SCRAM is not a good thing,
and if you had such a disparity, even at the
green/white level, that's a question to raise as to,
what does that mean to the public, what does that mean
to the licensee.
Loss of heat sink, I may be wrong, but I
looked at added up over a three-year period, and for
one of the plants it was a .7 transients per three-
year period. That means if you had one loss of heat
sink you would go into the white category. I don't
understand that. I think it shows the limitations of
risk-based versus risk-informed and how we have to
look at what we mean in terms of implementation here,
not just what does it mean in the risk model, but what
does it mean in the implementation world.
Also, a yellow for three SCRAMs, general
transient SCRAMs in a year, would not be very
appropriate to do.
Mitigating systems, the biggest issue we
are having now is unavailability. That is a real
problem. It's causing a lot of gnashing of teeth
among system engineers who have to go out and do the
Maintenance Rule one way, and the INPO indicator the
other way, and the ROP another way, and their PRA
person has a different way. So, we are working on
that, and I think a lot of the things that Steve is
working towards, reliability, not counting fault
exposure, are good things that we want to work, and we
really want to work on them faster.
CHAIRMAN APOSTOLAKIS: But, there is
resistance of changing these things and coming up with
a uniform set of definitions, which is a mystery to
me. I mean, this is the third time that I recall this
committee facing this issue, of what is reliability,
what is availability, and so on, and every time we
recommend that we need a White Paper with consistent
definitions, and every time we get something, we'll
think about it.
It's a very thinking agency here.
MR. HOUGHTON: Well, we got started with
the kick-off meeting amongst key players a month ago.
We have another meeting in May, and anything you can
do to support putting focus on this
CHAIRMAN APOSTOLAKIS: Well, I don't know
what to do. What should we do, Steve, to give focus?
Do you guys want to come before the committee? You
don't have to, you are industry, but I wrote a long
memo with four or five definitions, when was this
MR. HOUGHTON: Of course, we have to
CHAIRMAN APOSTOLAKIS: was it A.D. or
B.C., I can't remember.
MR. HOUGHTON: we have to satisfy a
number of interested parties. There's the Maintenance
Rule, which has its set of rules and the way it's been
doing things. You know, and that's a rule.
We've got PRA practitioners and they way
they look at things. We've got the INPO/WANO system,
okay, which they've been very good, in that they will
defer to the ROP definition, because they feel it's
more conservative. Okay. And, we have the ROP.
We have a basic underlying issue, which
is, in unavailability it's to be used to help
inspectors decide how much to inspect. Okay.
Inspectors inspect design-basis tech specs, allowed
outage times. The best definition we can come up with
is going to be more risk-based and oriented, okay, so
there's an important issue there that needs to be
addressed. So, it's not trivial to do this.
CHAIRMAN APOSTOLAKIS: I don't know what
the best way is. I mean, we'll leave it up to you how
you want to involve the committee. We can ask the
staff to come here, we can't ask you to come here.
MR. HOUGHTON: Well, I mean, we are happy
to come and talk and participate. I'm just
CHAIRMAN APOSTOLAKIS: It's something maybe
you can coordinate with Mr. Markley.
MR. HOUGHTON: It might be appropriate for
someone from the staff from NCR
CHAIRMAN APOSTOLAKIS: Will you be ready in
May when we have the full committee meeting to address
this issue, or is too soon?
MR. HOUGHTON: Well, we can lay out some
parameters of what we think the definition ought to
CHAIRMAN APOSTOLAKIS: Well, let's do that.
MR. HOUGHTON: Okay.
CHAIRMAN APOSTOLAKIS: In May, that will be
Friday, May 11th, at 2:30, we are discussing we have
an hour and a half on risk-based performance
indicators. Maybe that would be a good place to
MR. MAYS: George, let me interject a
little bit here if I may about that. We've been
working for a long time with the folks down in INPO
and EPIX stuff, and there was an NEI industry task
group to try to deal with this problem of different
data being collected different ways, to be reported to
five or six different entities, and what the
implications of that were. And, this is something
we've seen along the way.
What we found is that, the definition of
unavailability isn't so much the problem. The problem
tends to be the definition of the unavailability
indicator that you are using, because you know as well
as I do the unavailability definition from a classical
PRA or reliability definition is not that big of a
deal, but what happens is, when you start taking into
account other factors, such as, well, could the person
have restarted it or realigned it very quickly? You
take into account the factors, well, are we talking
about the automatic or the manual feature? You take
into account, well, are we talking about meeting its
design-based intent or its risk-significant intent?
There really isn't a real problem with getting the
amount of hours a piece of equipment is taken out of
service is not available, the performance function, we
have that data in various different varieties all over
the industry. The trick is to try to gather that
information sort of in its lowest common denominator
form and then create a kind of expert systems or smart
systems that will take that and use that information
to do the kinds of indicators that you want, depending
on who and what you want to look at.
For example, INPO wants to give credit to
plants that have more trains than they need to have
from a regulatory standpoint, so they can take those
trains out of service. So, they want to let them have
more unavailability. So, the way they do it is, they
define the unavailability indicator that doesn't
include those unavailable hours.
But, if you are doing a PRA, and you are
saying, what's the likelihood that these three trains,
instead of the two that are required, are going to
work or not, you need to know the unavailability of
all the three trains.
So, I found, and what I've seen, is that
the problem is not so much the definition of the
unavailability per se
CHAIRMAN APOSTOLAKIS: Unavailability,
though, there is a problem with the definition.
MR. MAYS: The problem we've seen, and I
think the problem we've run into in the Reactor
Oversight Process, is whether we are talking about
risk significant or design-basis function, whether we
are talking about auto or manual, whether we are
talking about how much credit you can take for being
able to realign or automatically resume.
MR. HOUGHTON: And, what system cascades to
MR. MAYS: Right.
And so, those are, in my opinion,
indicator definition problems, more so than
unavailability definition problems. And, what we've
seen is, I think, that there's a way to get to that
through common terms and definitions from a database
standpoint that will help a lot of these out.
CHAIRMAN APOSTOLAKIS: Well, the concept of
reliability is defined differently too, but, fine, I
mean, if we have a single document that explains all
these things, and comes up with a set of consistent
definitions, says that certain things are really
indicator problems rather than definitions, but right
now there isn't such a thing. So, I am all for it, to
develop something like that.
MR. HOUGHTON: As Steve was saying, there
is an industry consolidation group that's looking at
having a virtual database, which you can pluck
different data elements, common data elements from.
So, we are working that.
We are also working to meet again, I
think, about a week after your meeting with the key
players again from both industry and NRC, both
Maintenance Rule, PRA, ROP type people, so that we can
work towards these common data elements, so that we
don't have to waste our time fighting that.
CHAIRMAN APOSTOLAKIS: Okay, great. So,
you can brief us next time on the activities.
MR. HOUGHTON: Okay.
MR. BOYCE: NRR is also on that working
group, so that we also agree working towards common
definitions is the correct goal. In our most recent
public workshop for the Reactor Oversight Program last
week of March, that was one of the things we tried to
work towards, and we got a lot of input, but it's hard
to bring all those different organizations to a common
definition for many of the reasons that Steve just
said, they have different purposes for the use of the
data, but we are working on it, we are trying to get
CHAIRMAN APOSTOLAKIS: All right.
MR. HOUGHTON: Shutdown indicator, I think
it's a good effort, good start. However, at this
stage I don't think it passes the simple intuitive
capable of easy use that we need for a performance
indicator. It may be that it has greater value as a
Significance Determination Process, rather than as a
performance indicator per se.
I haven't had enough time to study the
details of it, but it looks a lot more difficult than
one would put on a public web site or that one would
CHAIRMAN APOSTOLAKIS: Which one is this
MR. HOUGHTON: The shutdown indicator.
CHAIRMAN APOSTOLAKIS: Oh.
MR. HOUGHTON: Let's see, I guess the last
Steve brought up, this is the first time I saw the
alternate approach. Certainly think on it, but I
think one of the principles we started with was, is
that aggregating the information to higher levels was
really counter to what the concept of doing the
performance indicators was, rather than have
aggregation to cornerstones or some higher level we
feel that the indicators ought to be as close to the
reality of what's going on the plant and be
actionable, such that we would say that having a SCRAM
indicator pass a threshold is actionable. You can
look at your SCRAM reduction program. Having a
particular system exceed a threshold allows you to go
focus first on that system and then do your root cause
and extent of condition, and look and see if it
applies elsewhere in the program.
So, we really feel that the program is
best left at a more granular level, in terms of
actionable level, in terms of performance indicators.
Now, that might mean a few more additional performance
indicators, we certainly would be willing to trade off
something workable in reliability as opposed to the
fault exposure, which has caused a lot of problems.
I guess my last point is, we look forward
to making the program better. We know it can be
better. I think, as I said a few minutes ago, a very
important part of these performance indicators is the
interface at the inspector level and how they view the
design basis versus the risk basis, which I think
Steve has talked also talked about, okay, which is
not it's not a trivial thing to change that mind
set, and it's the whole mind set in terms of all of
risk-based regulation versus the deterministic that we
CHAIRMAN APOSTOLAKIS: Thank you very much.
Maybe we can discuss now for a few minutes
what the presentation in May will consist of. Should
we go around the table and see what the members are
MR. LEITCH: I have a question for Steve,
if you don't mind.
CHAIRMAN APOSTOLAKIS: Sure.
MR. LEITCH: Just before we get into that.
I'm coming away with the impression that
the risk-based performance indicators are almost by
definition, by the criteria used to determine whether
you can establish a risk-based performance indicator,
they are almost by definition a lagging indicator, and
that most of the leading indicators you can't really
draw a distinct correlation between those indicators
I guess I thought you were going to tell
us at one point an example of a reactor that got into
trouble and going to try to back fit what the risk-
based performance indicators would look like and see
if it gives you any warning, any clue of impending
MR. MAYS: Those are both good questions.
Let me address the leading/lagging issue.
We tried to do that a little bit in the RBPI White
Paper discussion, and maybe we weren't as clear as we
need to be. The question, when you ask yourself about
leading and lagging indicators is leading and lagging
of what? I think you can see from the way we have
broken down risk from plant risk to the things
affecting containment, CDF and health effects, and
what are the things that affect CDF, I think you can
make the case that, for example, diesel generator
reliability, although that data is lagging of diesel
generator reliability, is leading of core damage
frequency, which is leading of public risk. So,
that's the perspective I have with respect to leading
Now, the issue about what are the causes
of those things to happen, I don't have really good
models right now to put in a risk perspective to say
the causes applied to reliability was getting worse or
availability was getting worse was this an aspect to
the way the plant is run, managed or operated? I
don't have that information. That would be even more
leading than what I have now.
But, I think we've made the case in the
White Paper that the combination of the fact that
we're looking at things that contribute to core damage
frequency, which contributes to public risk, makes
what we are doing leading, and, in fact, the
thresholds that we've chosen for those at the levels
we've chosen for them are significantly below the
existing public risk from all causes that relate to,
for example, early fatalities, that we have a pretty
good system of making sure we have a sufficient margin
built into the system so that even if we don't have it
completely down right we are not going to have gross
enough errors to really have a big impact on public
risk as compared to what we currently have for
accidental death rate.
So, from that standpoint I feel pretty
comfortable with the leading/lagging nature of what we
As we've shown here, if you want to get
more leading, or you want to hit higher level
indications, you have to do more aggregation and you
have to do more work of that nature.
The other issue, I think, with respect to
those, is that when you go back and look at how you
are getting the data and where you are setting the
thresholds, whether you are setting it at the input
point, or whether you are setting the thresholds at a
higher level, also affects what your leading or
lagging perception was.
I'm not sure I answered both of your
questions or not.
MR. LEITCH: Well, I guess the second one
had to do with, is there any evidence that if you use
the risk-based performance indicators, and tried
somehow to go back and back fit that to any of the
nasty events that we've had, is there any correlation
at all? And, I guess as long as it hasn't been core
MR. MAYS: That's one of those good
news/bad news things. The bad news is, is we can't go
back and relate this to actual core damage events, the
good news is, we can't go back and actually relate
this to core damage events.
MR. LEITCH: Yeah, right.
MR. MAYS: So, no, but one of the things
that we've looked at, and one of the things we've done
in the Reactor Oversight Process, was the question
becomes is, what constitutes poor performance, and
really when we were working on the ROP there really
wasn't a standard that you could compare against as to
what constitutes poor performance, other than things
like the watch list, or people who are on the INPO
trouble list or whatever. So, the ROP process went
back and looked at the current sets of indicators and
said, do these have good or reasonable correlation
with the plants that we have historically known to
have bad performance, and they had some data, and they
went back and did some analysis to say, these look
like they are reasonable, the bad performers tend to
fall out when we go back and look at the historical
The problem from risk-based performance
indicators is, I don't have data back into that realm
to make that I have two problems, one, I don't have
data on all these things back into the realms of the
1970s, '80s and early '90s, that I can compare these
to, to see whether they map out who were the "problem
plants," and I'm not even sure that the "problem
plants" were necessarily the worst ones from the risk
So, I have a problem on two levels. One
is the ground truth level and one is data to compare
MR. LEITCH: Yeah.
MR. MAYS: One of the things we did do, and
have done in looking at this, is we went back and
said, well, if the ROP was reasonable maybe we can go
back and take RBPIs over a similar period and look at
what the ROP did and see if we are coming up with
similar results or significantly different results, or
if we find differences do they make sense to us from
a risk perspective? And, that's what I meant by that
"face validity" comment.
DOCTOR KRESS: You still need to pick a
period you have the data for.
MR. MAYS: You need to pick a period where
you have comparable data for both processes, and the
best we can do right now is probably the '97 to '99
DOCTOR KRESS: Right.
MR. MAYS: We've taken a brief look at
that, and I think we found that we do a pretty
reasonable job of correlating with some of the stuff
that was in the ROP. We have more information that
they don't have, so you can't really compare what they
don't have to what we do have.
But, we did, we were able to go and look
and see where we found differences, and if it made
sense to us that the differences should exist, and the
kinds of things we found were, we found sometimes that
the RBPIs would have whites or yellows where the ROP
currently has greens or whites, and we looked at why.
And, when we looked at that, the most common reason
was because we were using plant-specific thresholds,
as opposed to generic or group thresholds.
We also found some cases where the ROP
would have whites or yellows, and we've had either
greens or whites, and we went back and looked at those
cases and what we found in those cases generally had
to do with things associated with the false exposure
time, when you take the false exposure time into
account in more of the way you would normally do it in
a risk assessment, and take into account the
reliability indicator portion, we found some of those
problems tended to go away.
But, we have gone back and looked at all
of those, and the other thing we found was the design
basis thing. If you were reporting unavailable
because it couldn't do its design basis function by
automatically starting, but was still capable of
manually starting, our indicators would indicate that
that was not a degradation as severe as the current
So, we've looked at that, but we haven't
published a formal side-by-side comparison like that,
and I'm not sure that there's anything we could do
anymore rigorously than a general comparison like that
in the first place.
Now, maybe if we were to go through this
and pilot some of these, what you would do is you
would run through the pilot with RBPI portions, and
you would run through and see what the comparison
would have been with the ROP, and then you go back and
ask yourself that "face validity" question again which
says, does it make sense that I'm having differences,
and do I believe that the differences are risk
significant? If you find that, you find that to be
something, as you said earlier, George, of benefit
that you want to do as a regulatory agency, then that
might be what you would do there. But, I think that's
part of looking at the stuff through the
DOCTOR KRESS: I hate to say this, because
it goes against my grain, but I think this is one of
those cases where your technical process itself is so
sound that I don't think you need to validate it
through real experience. I hate to say that, because
that's contrary to my usual belief.
MR. MAYS: I think you need to make the
case why what you have makes sense.
DOCTOR KRESS: Makes sense, it makes such
good sense, I don't think anymore validation than
you've already done is much worthwhile, because you
are validating against things that are not validated
themselves against reality.
MR. MAYS: It's a problem of where do you
find ground truth to compare it to.
DOCTOR KRESS: Yeah, so, you know, I
wouldn't search too much for more validation.
MR. MAYS: Well, we haven't done anymore
CHAIRMAN APOSTOLAKIS: Can we address the
issue of what to do?
DOCTOR KRESS: Of what to do?
CHAIRMAN APOSTOLAKIS: Yeah.
DOCTOR KRESS: Do you want to go around the
CHAIRMAN APOSTOLAKIS: Yeah, tell us if you
DOCTOR KRESS: Well, in the first place, I
think you need to tell us in general what the process
is, what you've done, but I would also be sure to get
to the three options that you talked about, because I
think it's very important that the full committee hear
I would talk about how you dealt with
shutdown, because it's significantly different than
the normal rest of the process, and I would go just a
little bit into the validation effort, comparing it to
the '97 data to '99 data, but not a lot. I wouldn't
spend a whole lot of time on that.
And then, I would point out this yeah,
I would point out this principle you are using,
progress versus perfection, and talk about things you
may improve in the future, because I think those are
questions that are going to come up.
So, that would be my opinion, George, on
what I think.
CHAIRMAN APOSTOLAKIS: Bob?
DOCTOR UHRIG: Well, I have the sense here
that what you are doing tends to validate the system
that is in place now. Am I stating that properly,
that you are getting comparable results to what you
are getting from the inspections that are going on
MR. MAYS: From the indicators that
DOCTOR UHRIG: yes.
MR. MAYS: I think we are getting
comparable readings in a number of areas. We are
getting more readings where they don't have readings
now, and where we have differences we know what the
basis for the differences are.
DOCTOR UHRIG: I think that should be
indicated, not on elaboration, but simply that's the
additional thing that I would add to what Tom has
CHAIRMAN APOSTOLAKIS: Graham?
MR. LEITCH: This last piece you covered
after lunch, the potential of the RBPIs went by me
awful fast. Frankly, I don't really understand what
was said there. I didn't have a chance to look at it
in advance, so I need some time to brush up on that,
but I think once more through that section, just a
little more slowly, might be helpful to the committee.
CHAIRMAN APOSTOLAKIS: Good.
DOCTOR BONACA: Yes, I pretty much agree
with the other points. Just a couple of things. One
is, you know, this is really a good effort, a good
visibility study of RBPIs, I mean, and to stress the
fact that, you know, the ROP is something different,
and, ultimately, there may be changes to that
depending on how well some of these RBPIs compare with
the existing ones.
The second point, the one that Graham
pointed to, it went very fast, and yet there is a lot
of merit on some of the alternatives, although I'm not
saying that they are going to be the likely one.
And, the third one is just a point I would
like to make, is that I think there is a more
systematic approach than it shows in the way we went
about this. I got the impression at the beginning
that you were saying, well, you know, whatever is
feasible we choose, and whatever cannot be done we
just don't go with it. I don't think you said that,
and I think that somehow I got a message, and maybe
you can communicate, that you have a systematic
approach. You are looking at containment, you are
looking at all the functions, and you do believe the
two that you could possibly identify there are already
significant of themselves and compare with the ones
you have right now whatever you have in the other side
program. I think that's important, because I didn't
get that message at the beginning.
CHAIRMAN APOSTOLAKIS: Okay, and we can
have another, I guess, of a little bit like you did
MR. BOYCE: It sounds like I'm on tap.
CHAIRMAN APOSTOLAKIS: Yeah.
MR. BOYCE: Can I just
CHAIRMAN APOSTOLAKIS: Maybe over some of
the issues that were raised regarding that memo.
MR. BOYCE: yes, I think comment number
seven is still down there, although I was still hoping
Steve had addressed your concern during the course of
CHAIRMAN APOSTOLAKIS: The full committee
probably needs to hear it.
MR. BOYCE: I was unsuccessful.
I did want to take, if I could, just a
second just to address some of the things that I heard
here, and give you a little bit bigger picture on, I
think, where NRR is coming from.
CHAIRMAN APOSTOLAKIS: Next time, not now.
MR. BOYCE: Well, I wanted to leave you
with just a general thought, if I could.
CHAIRMAN APOSTOLAKIS: Okay, go ahead.
MR. BOYCE: And, it relates to the tone in
that memo, as you pointed out, it was cool. The
approach I think we have is, is that the view we
have is that this project is ambitious, but it's
clearly in step with the Agency's direction to become
more risk informed, and so we support that, but we
have to be very, very cautious because we can't right
whole sail into this and then have some sort of
problem come up, like the SPAR models have a fatal
flaw, the licensees do not want to submit data to EPIX
anymore, and, therefore, the performance indicators
may not be valid anymore.
So, we are very conscious of the burden
that it places on licensees and the public acceptance
part of it. That's part of our performance goals, is
to enhance public confidence. And so, those sorts of
intangibles tend to get factored into technical
decisions on should we proceed with the risk-based PI
program, and that's why we are cautiously supportive
of this program.
CHAIRMAN APOSTOLAKIS: Okay, and that's
certainly something we want to discuss with the full
And, I also want to scrutinize Appendix F,
and discuss this issue of how the aleatory
uncertainties are handled, but other than that I think
we are in good shape. We had a good presentation
today, good discussion, we appreciate it. Thank you
very much, gentlemen, all of you.
DOCTOR KRESS: Once again, a very good
confident job and good presentation.
CHAIRMAN APOSTOLAKIS: Yes.
DOCTOR KRESS: Thank you very much.
CHAIRMAN APOSTOLAKIS: Nothing less is
expected of these guys.
DOCTOR KRESS: Yes, you know, we ought to
be raising the bar every time you guys come in,
CHAIRMAN APOSTOLAKIS: Yeah, it's over,
it's over, the subcommittee meeting is over.
(Whereupon, the above-entitled matter was
concluded at 1:55 p.m.)
Page Last Reviewed/Updated Tuesday, August 16, 2016