Protecting People and the EnvironmentUNITED STATES NUCLEAR REGULATORY COMMISSION
UNITED STATES OF AMERICA
NUCLEAR REGULATORY COMMISSION
ADVISORY COMMITTEE ON REACTOR SAFEGUARDS
***
MEETING: HUMAN FACTORS
U.S. NRC
Conference Room 28-1
Two White Flint North
11545 Rockville Pike
Rockville, Maryland
Friday, November 19, 1999
The committee met, pursuant to notice, at 8:30 a.m.
MEMBERS PRESENT:
GEORGE APOSTOLAKIS, Chairman, ACRS
DANA A. POWERS, Member, ACRS
THOMAS S. KRESS, Member, ACRS
JOHN J. BARTON, Member, ACRS
JOHN D. SIEBER, Member, ACRS
MARIO V. BONACA, Member, ACRS
ROBERT E. UHRIG, Member, ACRS
ROBERT L. SEALE, Member, ACRS. P R O C E E D I N G S
[8:30 a.m.]
DR. APOSTOLAKIS: The meeting will now come to order.
This is a meeting of the ACRS Subcommittee on Human Factors.
I'm George Apostolakis, chairman of the subcommittee. The ACRS members
in attendance are Mario Bonaca, John Barton, Robert Seale, Dana Powers,
Jack Sieber and Tom Kress.
The purpose of this meeting is to review a proposed revision
to NUREG 1624, Technical Basis and Implementation for a Technique for
Human Event Analysis, ATHEANA, period and assist staff research
activities related to human reliability analysis, pilot application of
ATHEANA to assess design basis accidents and associated matters. The
subcommittee will gather information, analyze relevant issues and facts
and formulate proposed positions and actions as appropriate for
deliberation by the full committee.
Mr. Juan Piralta 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 October 14, 1999. The transcript of this
meeting is being kept and will be made available as stated in the
Federal Register notice. It is requested that speakers first identify
themselves and speak with sufficient clarity and volume so that they can
be readily heard.
We have received no written comments or requests for time to
make oral statements from members of the public. We have to recess at
11:45, because I have to go to another meeting, and then, we will
reconvene again at maybe 12:45, okay? So, if you can plan your
presentation around that schedule, that will be good.
We will now proceed with the meeting, and I call upon Mr.
Mark Cunningham, for a change, to begin.
[Laughter.]
DR. APOSTOLAKIS: Was there ever a meeting where Mr.
Cunningham was not the first speaker?
[Laughter.]
DR. APOSTOLAKIS: We all ask.
MR. CUNNINGHAM: Probably one or two in the last 20 years;
not much beyond that, it seems.
Good morning.
DR. APOSTOLAKIS: And Dr. Uhrig just joined us, for the
record.
MR. CUNNINGHAM: All right; on the agenda, I've got a couple
of items to begin with this morning. First is just an overview of what
we're doing. The second is topics related to international efforts.
I'd like to put the international efforts, to delay that a little bit
and discuss it after the ATHEANA presentation, because I think the
context is much better after you've heard more about ATHEANA and the way
we're treating human errors and things, unsafe acts, I'm sorry, that
sort of thing.
But anyway, by introduction, we have I guess by and large
one big topic and a couple of smaller topics to discuss this morning.
The big topic is the work we've been doing over the last year or so to
the ATHEANA project to respond to the peer review that we had in Seattle
awhile back, June, okay? That's the main topic for the day, so we'll
talk about that; we'll talk about the structure of ATHEANA, what the
objectives of the project are and then have an example.
One of the things we've been doing over the last year is
demonstrating the model in an analysis of a fire accident scenario in a
plant that gets involved with this self-induced station blackout, a
SISBO plant, if you will. After that, we'll come back and talk about
two smaller topics. One is a base proposal, which are basically our
international efforts in the human reliability analysis. We had some
work underway for the last couple of years with CSNIs, PWG-5, Principal
Working Group 5, and you had errors of commission; we also had a CUPRA
program related to trying to relate risk -- bring into risk analysis the
impact of organizational influences. So, I'll talk briefly about those
later on in the morning or right after lunch or something like that.
DR. APOSTOLAKIS: Oh, I forgot to mention that Mr. Sorenson,
a fellow of the ACRS, will make a presentation on safety culture after
lunch, and we would appreciate it if some of you guys stay around and
express comments and views. This is an initiative of the ACRS, and
certainly, your views and input would be greatly appreciated. So don't
disappear after the ATHEANA presentation.
MR. CUNNINGHAM: We won't. Most of us won't.
DR. APOSTOLAKIS: Good.
MR. CUNNINGHAM: With that, I'll turn it over to Katharine
Thompson. Katharine is the project manager of the ATHEANA project in
the office built by two support people, John Forester from Sandia and
Alan Kolaczkowski from SAIC. We've got some others in the audience,
too, but we'll get back to that in a minute.
DR. THOMPSON: Good morning, and it's my pleasure to be here
this morning to discuss ATHEANA with you for the first time, I guess. I
know you've heard a lot about it.
DR. APOSTOLAKIS: We should invite you more often,
Katharine.
[Laughter.]
DR. POWERS: Well, George, I will point out that the first
speaker before the committee usually gets asked a fairly similar
question.
DR. APOSTOLAKIS: Yes; go ahead, Dana.
DR. POWERS: What in the world qualifies you to speak before
this august body?
[Laughter.]
DR. THOMPSON: I have orders from my manager.
[Laughter.]
DR. POWERS: No, I'm serious; could you give us a little bit
of your background?
DR. THOMPSON: Oh, sorry; I have a Ph.D. in industrial and
organizational psychology. I've been at the NRC for about 10 years. I
was in NRR and human factors for a few years, and then, I went as a
project manager for the Palo Verde plant. I've been over here in the
research and assessment branch for about 5 or 6 years, and I've been
working on ATHEANA for the past about 5 years.
DR. POWERS: What in the world makes you think that this
body will understand anything you have to say?
[Laughter.]
SPEAKER: We'll be slow in delivery.
DR. THOMPSON: Okay; just a brief outline of the
presentation. I'm going to be discussing the overview and a brief
introduction. Dr. John Forester will be going through the structure of
ATHEANA and how it's done. Alan Kolaczkowski will be talking about the
fire application, and then, I'll be back to talk about some conclusions
and some follow-up activities.
We're not going to talk about the peer review in the
interests of time, but in the back of your handout, you have all of the
slides and discussion of the peer review, so you can look at that in
your own time.
DR. APOSTOLAKIS: Unless we raise some issues.
DR. THOMPSON: Unless you raise some issues.
I guess the first question that always comes is why do we
need a new HRM method? And so, we've talked about this and looked at
accidents that happen in the industry and other industries, events that
have happened, and certain patterns and things come to the surface.
What we're finding is that a lot of problems involve situation
assessment; that scenarios and the events deviate from the operator's
expectation. Perhaps they were trained in one way on how to approach a
situation, and the scenario didn't happen that they were trained on.
We've seen that plant behavior is often not understood, that
multiple failures happen that are outside the expectations of the
operators, and they don't know how to respond to this or how to handle
it properly. They weren't trained on how to follow these scenarios.
And we also know that plant conditions are not addressed by procedures.
A lot of times, these things don't match. The procedures tell them how
to go through a scenario, but yet, the scenario isn't matched with the
procedures at hand, so that they may do something that's not in the
procedures; that could, in fact, worsen the conditions.
And these types of things aren't handled appropriately in
current ERAs and HRAs, and so, we need to address these problems with
situation assessment and how the plan is understood by the operators.
DR. APOSTOLAKIS: Now, this thing about the procedures is
interesting. Isn't it true that this agency requires verbatim
compliance with procedures, unlike the French, for example, who consider
them general guidelines?
DR. POWERS: Guidance.
DR. APOSTOLAKIS: Yes; it's like traffic lights somewhere
else.
So how -- what are we going to do with this? I mean, should
the agency change its policy?
MR. CUNNINGHAM: We are probably not the best people to say,
but I don't think that's the policy of the agency, to follow -- require
verbatim compliance with the procedures.
DR. BARTON: George, the agency requires that you have to
procedures to conduct operations, to handle emergencies, et cetera.
Some procedures are categorized in different categories: continuous
use, reference, stuff like that. But there really isn't --
DR. APOSTOLAKIS: There is no --
DR. BARTON: -- a requirement that you do verbatim
compliance.
DR. BONACA: Although the utilities --
DR. BARTON: Utilities have placed compliance, strict
compliance, on certain groups of procedures, and they have also policies
that say if you can't comply with the procedure, what you do: stop and
ask your supervisor, change the procedure, et cetera. But I don't think
there are any regulations that say you have to follow procedures
verbatim.
DR. APOSTOLAKIS: Although we have been told otherwise,
though. What's that?
DR. BONACA: Control room procedures, however, in an
emergency, EOPs, for example, there is following verbatim to line by
line.
DR. APOSTOLAKIS: But these are the ones that Katie is
talking about, right? EOPs?
DR. BONACA: Yes.
DR. APOSTOLAKIS: Not procedures for maintenance. I mean,
you're talking about --
MR. CUNNINGHAM: Again, I don't know that it's a requirement
of the agency that they follow line-by-line the procedures. It's my
understanding that it's not.
DR. UHRIG: It was 20 years ago at one point, but that was
believed.
MR. CUNNINGHAM: Okay.
DR. BONACA: Well, the order in which you step through an
emergency procedure is very strict. I mean, at least -- I don't know if
it is coming from a regulation, but it is extremely strict. You cannot
-- I mean, the order of the steps you have to take; that's why you have
the approach in the control room with three people, and one reads the
procedure; the others follow the steps.
DR. APOSTOLAKIS: Yes, that's true.
MR. CUNNINGHAM: Again, I think all of that is very true. I
just don't think it's a requirement -- it's not in the regulations that
they do that is my understanding.
DR. APOSTOLAKIS: You say you are not the appropriate
people. Who are the appropriate people who should be notified?
MR. CUNNINGHAM: I'm sorry?
DR. APOSTOLAKIS: Maybe that will do something about it.
MR. CUNNINGHAM: Who --
DR. APOSTOLAKIS: Who in the agency is in charge of the
procedures and compliance?
MR. CUNNINGHAM: It's our colleagues in NRR, obviously, and
where exactly in the last reorganization this ended up, I'm not quite
sure.
DR. APOSTOLAKIS: Okay.
MR. CUNNINGHAM: But the issue of whether or not there is
verbatim compliance is an NRR issue that --
DR. SEALE: It might be interesting to discuss this with
some inspectors in the plant.
DR. KRESS: Whenever we've heard -- one of these things that
always seems to show up. I'm sorry; I can't talk to them and listen at
the same time, but it seems to me like there was almost an implied -- on
these procedures.
DR. APOSTOLAKIS: Yes.
DR. KRESS: Whether it's real or within the regulations or
not.
DR. BONACA: But that certainly has been interpreted by now
by the licensees. I mean, for the past 10 years, especially -- even the
severe accident guidelines, in some cases, where you look at the
procedures, they are very strictly proceduralized, I mean. And you
check to see that people do not even in the simulator room do not invert
the order of the stuff.
DR. THOMPSON: Yes, but a lot of that came from the
analysis, because following the procedure requirement, it's the next
step that you must deal with, I don't ever recall a regulation requiring
verbatim compliance. We had company policy about certain procedures.
DR. APOSTOLAKIS: Okay.
DR. THOMPSON: Okay; so what we know from all of these
reviews of accidents and events is that situations in the context
creates the appearance that a certain operator action is needed when, in
fact, it may not be and that operators act rationally; they want to do
the right thing; they try to do the right thing, and sometimes, the
action is not the appropriate action to take. The purpose for ATHEANA,
then, is to provide a workable, realistic way to identify and quantify
errors of commission and errors of omission.
There are three objectives of ATHEANA. First is to enhance
how human behavior is represented in accidents and near miss events. We
do this by looking at the decision process involved, how people -- their
information processing abilities and how they assess a situation, and we
also integrate knowledge from different disciplines. We look -- we have
technology factors, engineering risk assessments. We try to incorporate
many different areas of knowledge there.
DR. POWERS: I guess I'm struck by how this view graph would
have been written by somebody -- who developed human error analysis
methodologies they use now. They probably would use this view graph and
just change the title, right? Everybody that advances the human -- our
reliability analysis program says he's going to make it realistic; he's
going to integrate perspectives of ERA with plant engineering,
operations training, psychology, risk-informed and have insights. I
mean, this is true of any conceivable human error analysis.
MR. CUNNINGHAM: In theory. Now, we could go back perhaps
in another session and talk about how much did other methods really
accomplish this, and I think what you see, and you hear stories of how,
in the poorer qualities HRAs, if you will, how this is implemented in a
way that, in fact, the issues such as psychology and operations and
training and things like that are handled on a rather -- one way to put
it is a crude way, and one way would be just a mechanical way or
something like that.
DR. POWERS: You know, I mean when you look for things like
your hallowed Navier Stokes equations, people come up with --
DR. KRESS: Hallowed, not hollowed.
DR. POWERS: That's right, hollowed.
[Laughter.]
DR. POWERS: The fount of all wisdom, and you call it the
big bang; everything else was just thermohydraulics.
[Laughter.]
DR. SEALE: And a little chaos thrown in.
[Laughter.]
DR. POWERS: You know, in your equations, you say, well,
we'll make an approximation. We may have zeroeth Ns, and you can see
that there is no dimensionality in the zeroeth approximation, and then,
you have first order ones and second order ones and third order ones,
and it's very clear when somebody is getting more realistic and
incorporating more terms. How am I going to look and see that this
ATHEANA program is more realistic? You know, what is it that says
clearly that this is more realistic than what was done many, many years
ago for the weapons programs?
MR. CUNNINGHAM: I guess in my mind, there would be a couple
of clues. I guess one would be how well we can mimic, if you will, or
reproduce the real world accidents that Katharine started talking about,
and again, those are the accidents that are, if you will, I think of
them as the more catastrophic accidents. If you look back and see,
investigate human performance in catastrophic accidents, how well does
this model -- I don't want to say predict but work with those types of
events?
DR. KRESS: You're not talking about neutral.
MR. CUNNINGHAM: No, I'm talking about in general. I can
think of --
DR. KRESS: Can you transfer that technology to technology?
MR. CUNNINGHAM: Yes, I think you can, and that's kind of
one of the subtle, underlying presumptions is that the human performance
in catastrophic accidents can be translated across different industries,
highly complex, high-tech industries, if you will: aircraft, chemical
facilities and that sort of thing.
DR. APOSTOLAKIS: I think there is a message here,
Katharine: use your judgment as you go along, and skip the view graphs
that are sort of general and focus on ATHEANA only. Do not raise
anything until you come to the specifics. Otherwise, you're going to
get discussions like this.
[Laughter.]
DR. APOSTOLAKIS: So can you go on, and we'll come back to
these questions?
DR. THOMPSON: Skip the next one, John.
This is just to show you the basic framework of ATHEANA and
to underscore again -- well, we use different ones here; that's the left
part. Psychology, engineering -- this is something we've been working
on. The left-hand side shows you the elements of psychology, human
factors engineering that are folded into the framework.
DR. APOSTOLAKIS: Go ahead.
DR. THOMPSON: And then, it flows into the PRA logic models
and the rest. You've seen this before. John is going to talk more
about this in the future, so I don't want to spend too much time on this
right now.
DR. APOSTOLAKIS: I have a couple of comments.
DR. THOMPSON: Okay.
DR. APOSTOLAKIS: I have complained in the past that
error-forcing context is a misnomer, and then, I read your chapter 10,
which tells me that there may be situations where the error-forcing
context really doesn't do anything. So I don't know why it's forcing.
I notice that some of the reviewers also said that it's probably better
to call it error-producing, error -- I don't know, some other word than
forcing, because you, yourselves say in chapter 10 that the probability
of error, given an error-forcing context, is not one, may not be one.
DR. THOMPSON: Right.
DR. APOSTOLAKIS: Second, I don't understand why you call
them unsafe actions. I fully agree that the human failure event makes
sense, but until you go to the human failure event, you don't know that
the action is unsafe. I mean, you insist -- in fact, you just told us
-- that people aren't rational, and I'm willing to accept that. So the
poor guy there took action according to the context, which led to a
human failure event. So I don't think you should call it unsafe. I
mean, human actions -- don't you think that that would be a better
terminology?
And then, finally, coming back to Dr. Powers' question, I
give you my overall impression of the report. I think the greatest
contribution that ATHEANA has is the extreme attention it paid to the
plant, at the plant conditions; that there is an awfully good discussion
of how the plant conditions shape the context. But I must say that
chapter 10 was a disappointment. The quantification part, I didn't see
anything there that really built on the beautiful stuff that was in the
previous chapters. In fact, it just tells you go find a method and use
it.
It's a little harsh, but, I mean, in essence, that's what it
says. I mean, I have this context. I spent all this effort to find the
error-forcing context. And then, all you are telling me is now, you can
use half. You can use, you know, slim model if you like. I thought I
was going to see much more. I mean, this thing of error mechanisms has
always intrigued me, why you bother to use it. And then, in chapter 10,
you don't use it, which is sort of what I expected. I mean, I can't
imagine anybody quantifying error mechanisms.
So I don't know if this is the proper place to discuss this,
because it's jumping way ahead, but I'm just letting you know that
chapter 10, I thought, was a let-down after the wonderful stuff that was
in the previous chapters.
MR. CUNNINGHAM: Yes, I think we are getting a little ahead
of --
DR. APOSTOLAKIS: Yes, okay.
MR. CUNNINGHAM: I mean, after John and Alan talk for
awhile, we can come back to this.
DR. APOSTOLAKIS: But, I mean, one part of the answer to Dr.
Powers is that this is really the first HRA approach that really paid
serious attention to the plant conditions, and I think that is very,
very good, very good, but we are really -- we are not just speculating
now. You guys went out of your way to see how this circle there, plant
design, operations and maintenance and plant conditions shape the
context. I've always had reservations about the error mechanisms, but I
deferred to people more knowledgeable than I.
But chapter 10 now makes me wonder again. So, but the
terminology, I think, is very important. I'm not sure that you should
insist calling it error-forcing context when you say in chapter 10 that
-- I don't remember the exact words but, you know, sometimes, you know,
it doesn't really matter. How can it be forcing it?
Yes, John?
MR. FORESTER: Do you want me to comment on it?
DR. APOSTOLAKIS: I want you to comment on this.
MR. FORESTER: I suggest we come back and --
DR. APOSTOLAKIS: Great.
MR. FORESTER: -- the natural progression of the talk will
get us to chapter 10.
DR. APOSTOLAKIS: Okay; fine, fine.
MR. FORESTER: Sometime today so --
DR. APOSTOLAKIS: Do you have any reaction to the comments
on the terminology? I mean, last time, you dismissed me. Are you still
dismissing me?
[Laughter.]
DR. THOMPSON: We'll come back to it.
MR. FORESTER: We will come back to it.
MR. KOLACZKOWSKI: The answer is yes.
DR. APOSTOLAKIS: Well, then, that gives me time to find
your exact words in chapter 10.
[Laughter.]
DR. APOSTOLAKIS: Okay.
DR. THOMPSON: This slide going real fast. I wanted to just
briefly recognize the team, because they all did a wonderful, wonderful
job, and it, again, underscores the different disciplines we've brought
to this program. We've got psychologists, the first three,
specifically.
DR. APOSTOLAKIS: Always pleased to see names that are more
difficult to pronounce than my own.
[Laughter.]
MR. KOLACZKOWSKI: I don't see any such names here.
[Laughter.]
DR. THOMPSON: He's referred to as Alan K., because I can't
pronounce it either.
[Laughter.]
DR. THOMPSON: Engineers, risk assessment experts,
psychologists, human factors, so we've brought all of the disciplines to
this project that we need.
DR. APOSTOLAKIS: By the way, I hope you don't misunderstand
my comments. I really want this project to succeed, okay? So I think,
you know, being frank and up front is the best policy. So I must tell
you that it was not a happy time for me when I read chapter 10.
MR. CUNNINGHAM: We appreciate that over the years, we've
gotten a lot of good advice from the various subcommittees and
committees here, and we appreciate that and take it in that vein, even
though we may take your name in vain occasionally.
[Laughter.]
DR. POWERS: We are probably in good company.
DR. APOSTOLAKIS: Now, you know why Mr. Cunningham is always
there --
[Laughter.]
DR. APOSTOLAKIS: -- every time we meet. He knows how to
handle situations like this.
[Laughter.]
MR. FORESTER: Yes; I am John Forester with Sandia National
Laboratories, and I'm, I guess, the project manager, the program
manager. I work for Katharine, and I'm the project leader for the team.
DR. APOSTOLAKIS: She's not Kitty anymore? Is it Katharine
now?
MR. FORESTER: Katharine, yes.
DR. APOSTOLAKIS: Okay.
[Laughter.]
MR. FORESTER: For this part of the presentation, I'm going
to discuss the structure of ATHEANA, and what I'd like to do is focus on
the critical aspects and processes that make up the ATHEANA method.
DR. APOSTOLAKIS: So, you skipped the project studies.
DR. THOMPSON: I'm sorry; I'll get back to that at the end
when we talk about the completion.
DR. APOSTOLAKIS: Okay.
MR. FORESTER: Okay; ATHEANA includes both a process for
doing retrospective analysis of existing events and a process for doing
prospective analysis of events.
DR. KRESS: A retrospective? Is that an attempt to find out
the cause?
MR. FORESTER: Right, an analysis of the event to find out
what the causes were and, you know, ATHEANA has had a process or a
structure, at least, for doing that for quite awhile, to be able to
analyze and represent events from the ATHEANA perspective so that you
can understand what the causes were and also, by doing that in this kind
of formal way, you'd have a way to maybe identify how to, you know, fix
the problems in a better way.
DR. KRESS: And you can use that retrospective iteratively
to improve some of the models in the ATHEANA process?
MR. FORESTER: Yes; you know, the idea was that by doing
these retrospective analyses, we learn a lot about the nature of events
that had occurred and then can take that forward and use it in the
prospective analysis.
DR. APOSTOLAKIS: But today, you will focus on prospective
analysis.
MR. FORESTER: That is correct; yes, I just want to note
that one of the recommendations from the peer review in June of 1998 was
that we had the structure for doing the retrospective, but we did not
have an explicitly documented process for doing the retrospective, and
we have included that now, okay? And we do see that as an important
part of the ATHEANA process in the sense that, you know, when plants or
individuals go to apply the process, they can look at events that have
occurred in their own plant and get an understanding of what the kinds
of things ATHEANA is looking for, sort of the objectives of it, and that
way, it will help them be able to use the method, in addition to just
learning about events in the plant and maybe ways to improve the process
or improve the problem, fix the problem.
Okay; now, we do see in terms of the prospective analysis,
as George said, we're going to focus on that mostly today. We do see
the process as being a tool for addressing and resolving issues. Now,
those issues can be fairly broadly defined in the sense of we're going
to do an HRA to support a new PRA, but we also see it as a tool to use
more specifically in the sense -- for example, you might want to extend
an existing PRA or HRA to address a new issue of concern; for example,
maybe, you know, the impact of cable aging or operator contributions to
pressurized thermal shock kind of scenarios or fire scenarios. So it
can be used in a very effective manner, I think, to address specific
issues.
Also, maybe, to enhance an existing HRA or, you know,
upgrade an existing HRA to be able to -- for purposes of risk-informed
regulation submittals and things like that. So it can be a very
issue-driven kind of process.
The four items there on the bottom are essentially sort of
the major aspects of the tool, and I'm going to talk about each one of
those in detail, but in general, the process involves identifying base
case scenarios; sort of what's the expected scenario given a particular
initiator and then trying to identify deviations from that base case
that could cause problems for the operators.
Another major aspect of the --
DR. KRESS: Are those the usual scenarios in a PRA that
you're talking about?
MR. FORESTER: The -- well, no, the base case is sort of --
I'll go into more detail about what the base case scenario actually is,
but it is what the operators expect to occur, and it's also based on
detailed plant engineering models, okay? So maybe you'll lift something
from the plant FSAR, but I'll talk about that a little bit more.
And again, another major aspect of the revised method is
that we try to clarify the relationship between the deviations, the
plant conditions and the impact on human error mechanisms and
performance shaping factors. So we tried to tie that together a little
better, and I think we've created at least a useful tool to do that
with. And then, finally, the other major aspect is the integrated
recovery analysis and quantification, and I would like to say Kitty has
already pointed out that I'll kind of go through the general aspects of
the process, and then, Alan is going to give us an illustration of that
process, okay?
[Pause.]
MR. FORESTER: Okay; I think as we mentioned earlier, sort
of the underlying basis for the prospective analysis is that most
serious accidents occur when the crew sort of gets into a situation
where they don't understand what's going on in the plant.
DR. APOSTOLAKIS: Is this Rasmussen's knowledge-based
failure?
MR. FORESTER: Yes, I guess it would be. It's where the
procedures don't maybe fit exactly; they may be technically correct, but
they may not fit exactly, and, well, even in the aviation industry or
any other kind of industry, what you see in these kind of serious
accidents was that they just didn't understand what was going on.
Either they couldn't interpret the information correctly. I mean, in
principle, I guess it could have been responded to in a rule-based kind
of way, but they didn't recognize that, so it did put them into a
knowledge based kind of situation.
DR. KRESS: When I read that first bullet, I'm thinking of
nuclear plants because it comes from the broad plan.
MR. FORESTER: Yes; that's true, but there have been some
events. I mean, they haven't led to serious events, necessarily, and
even beyond TMI and --
DR. KRESS: Yes, but that's one data point.
MR. FORESTER: I mean, there are other events, though, that
haven't gone to core damage or, I mean, that haven't really led to any
serious effects.
DR. KRESS: But you're getting this information from --
MR. FORESTER: Yes, yes; okay.
DR. KRESS: Because in designing nuclear plants, we talk
about conditions not understood. We've gone to great pains to get that
out. I'm sorry; I'll just quit talking.
[Laughter.]
MR. FORESTER: It does seem, even in the nuclear industry,
you know, there are times where people do things wrong. I mean, it
doesn't lead to serious problems, but people do, you know, they bypass
SPASS --
DR. SEALE: You know, it really goes back to George's
comment about human error. Human error is a slippery slope. It's not a
cliff. And, in fact, when human error occurs, the angle of that slope
will vary from error to error, and while you may talk about TMI as a
case where you led to an accident, I bet you you could find a dozen
where people did something, recognized that they were on a slippery
slope, and recovered, and that seems to me, that should be just as
useful an analysis, an identification to do in your ATHEANA process as
was the TMI event, because it's the process you're trying to understand.
MR. CUNNINGHAM: No, I think that's right; you learn from
your mistakes. You also learn from the mistakes you avoid.
DR. SEALE: And the ability to recover is important
knowledge.
MR. CUNNINGHAM: Yes; there's a lot of work that's been done
about TMI; an operator response to initial events, and as you said,
there is still the residual that they don't understand, and that's where
we can get into very severe accidents, even after all that training.
DR. POWERS: It seems to me that a double-ended guillotine
pipe break, that's a severe accident that a crew would understand
absolutely what it was doing in a double-ended guillotine pipe break.
DR. KRESS: So we are never going to have one.
[Laughter.]
DR. POWERS: If we had one, you would damn well know what
happened.
[Laughter.]
DR. POWERS: You wouldn't be able to mistake it for much.
It seems like what you're saying may be true for accidents that are of
real concern to us, but it's going to run counter to the DBAs. The
DBAs, you know what's going on, and it doesn't seem like it applies to
the DBAs.
MR. CUNNINGHAM: DBAs are obviously very stylized accidents.
DBAs themselves are very stylized accidents, and the training, you know,
25 years ago was fairly stylized to go with those accidents. We've made
a lot of progress since then in taking a step back from the very
stylized type of approach, but you can still have accidents or events.
The one that comes to mind for me is the Rancho Seco event of -- I don't
know -- the early eighties or something like that, where they lost a
great deal of their indication; another indication was confusing and
that sort of thing. It's not a design-basis accident, but it was a
serious challenge to the core, if you will.
DR. APOSTOLAKIS: Isn't, John, I don't see anything about
the values of operators, the references; again, the classic example is
Davis-Bessie, you know, where the guy was very reluctant to go to bleed
and feed and waited until that pump was fixed, and the NRC staff, in its
augmented inspection team report, blamed the operators that they put
economics ahead of safety. The operators, of course, denied it. The
guy said, no, I knew that the pump was going to be fixed, but isn't that
really an issue of values, of limits? It's a decision making problem.
MR. CUNNINGHAM: Right.
DR. APOSTOLAKIS: Where in this structure would these things
-- are these things accounted for? Is it in the performance shaping
factors, or is it something else?
MR. FORESTER: Well, one place it comes through is with the
informal rules. We try to evaluate informal rules. And if there's sort
of a rule of, you know, we've got to watch out for economics, I mean, in
their minds, it may not be an explicit rule, but in their minds, they're
not going to do anything that's going to cost the utility a lot of
money. That's one way we try to capture it.
There's also -- we try and look at their action tendencies.
We have some basic tables in there that addresses both the BWR and PWR
operator action tendencies, what they're likely to do in given
scenarios.
DR. APOSTOLAKIS: But if I look at your multidisciplinary
framework picture that you showed earlier, I don't see anything about
rules. So the question is where, in which box, you put things like
that.
MR. FORESTER: Well, I guess it would probably be sort of
part of the performance shaping factors.
DR. APOSTOLAKIS: I'm sorry, what?
MR. FORESTER: Well, overall, the impact of rules would sort
of be -- or of what you're describing here, and I used informal rules as
how we get at that in terms of the framework, it would certainly be
covered under part of the error forcing context, essentially.
DR. APOSTOLAKIS: But this is the performance shaping
factor, part of the performance shaping factor?
MR. FORESTER: I think it -- I guess it would also -- I'm
not sure we'd directly consider it as a performance shaping factor.
DR. APOSTOLAKIS: What is a performance shaping factor in
this context? Give us a definition.
MR. FORESTER: Well, procedures, training, all of those
things would be -- the man-machine interface, all those would be --
DR. APOSTOLAKIS: Technological conditions? Is that
performance-shaping factors?
MR. FORESTER: Stress and --
DR. APOSTOLAKIS: So the error forcing context is the union
of the performance shaping factors and the plant conditions. Is that
the correct interpretation of this?
MR. FORESTER: That's a correct interpretation.
DR. APOSTOLAKIS: So clearly, values cannot be part of the
plant conditions, so they must be performance-shaping factors. I mean,
if it's the union --
MR. KOLACZKOWSKI: I'm Alan Kolaczkowski with SAIC. Yes, if
you want to parcel it out, if you want to actually put tendencies of
operators or roles into a box, it would best fit in the performance
shaping factors, yes, but the reason why I think we're struggling is
that we recognize that to really define the error-forcing context, you
have to think about the plant conditions and all the influences on the
operator in an integrated fashion, and it's hard to parcel it out, but
if you want to put it in a box, I would say yes, it's affecting the
performance shaping factors.
DR. APOSTOLAKIS: That's what the box says: all of these
influences --
MR. KOLACZKOWSKI: I understand.
DR. APOSTOLAKIS: -- are the PSFs, because there's nothing
else.
MR. FORESTER: Well, it could be more specified, I would
say, in the sense that part of what you're bringing up is augmented in
the organizational factors, maybe even team issues, things like that,
which are going to be -- which are certainly going to contribute to the
potential for error. Those are not explicitly captured. In some sense,
they could be looked at as part of the plant conditions, and they could
also be looked at as performance shaping factors.
DR. APOSTOLAKIS: Now, this sector on the left, what do you
mean by operations?
MR. FORESTER: Just the way they do things there, the
procedures, their modus operandi, I guess, as to the way they run the
plant.
DR. APOSTOLAKIS: Is what other people call safety culture
there?
MR. FORESTER: I think that's more --
DR. APOSTOLAKIS: No, but that's part of it. there's an
error there on the left, plant design, operations and maintenance. I
remember the figure from Jim Reason's book, where he talks about line
management deficiencies and valuable decisions. Are you lumping those
into that circle, or are you ignoring them? I mean, the issue of
culture --
MR. FORESTER: We have not explicitly tried to represent
those yet.
DR. APOSTOLAKIS: But this is a generic figure, so that's
where it would belong, right?
MR. FORESTER: I'm not sure I would normally necessarily
pigeonhole it there. It's all part of that whole -- the whole error
force in context and what feeds into the error force in context.
DR. APOSTOLAKIS: But the error force in context is shaped
by these outside influences. It does not exist by itself. You have
these arrows there.
MR. FORESTER: Right.
DR. APOSTOLAKIS: So this is an outside influence, so, for
example, if I wanted to study the impact of electricity market
deregulation, that would be an external input --
MR. FORESTER: Yes.
DR. APOSTOLAKIS: -- that would affect the performance
shaping factors and possibly the plant condition.
MR. FORESTER: Yes; that is correct.
DR. APOSTOLAKIS: Okay.
MR. CUNNINGHAM: That is correct.
DR. APOSTOLAKIS: So all of these are external influences
that shape what you call error force in context.
MR. CUNNINGHAM: That's right. This is a very conceptual
description of the process.
DR. APOSTOLAKIS: Yes.
MR. CUNNINGHAM: And it's probably a little broader than
ATHEANA is today, but again, if we could go back and get into ATHEANA as
it is today, it might help --
DR. APOSTOLAKIS: Okay.
MR. CUNNINGHAM: -- some of the others understand what we're
going through here.
MR. FORESTER: Well, given what we've identified as the
nature of serious accidents, we think a good HRA method should identify
these conditions prospectively, and we have several processes that we
use to do that. Mr. Chairman, I'm going to talk about these in more
detail, to identify the base case scenarios, and again, these are
conditions that are expected by the operators and trainers given a
particular initiating event.
They may want to identify potential operational
vulnerabilities, and these might include operators' expectations about
how they think the event is going to evolve. It could include
vulnerabilities and procedures; for example, where the timing of the
event is a little bit different than what they expect. The procedure
could be technically correct, but there could be some areas of ambiguity
or confusion possibly.
And then, based on those vulnerabilities, at least part of
what we use is those vulnerabilities, then try and identify reasonable
deviations from these base case conditions, to sort of see if there are
kinds of scenarios that could capitalize on those vulnerabilities and
then get the operators in trouble.
DR. APOSTOLAKIS: So I think it's important to ask at this
point: what were the objectives of the thing? It's clear to me from
the way the report is structured and the way you are making the
presentation that the objective was not just to support PRA.
MR. FORESTER: Not just to support PRA, no; I guess that's
maybe how we started out, but I think the method itself can be used more
generally than in PRA. I think it needs to be tied to PRA because of
some of the ways we do things, but no, certainly, it could be used more
generally.
DR. APOSTOLAKIS: What other uses do you see?
MR. FORESTER: You can do qualitative kind of analysis, so
if you're not doing a PRA, you don't need explicit quantitative
analysis. So with, for example, in the aviation industry, there is not
a whole lot of risk assessment done as far as I know on what goes on in
the airplane cockpits, but that doesn't mean that you couldn't use this
kind of approach to develop interesting scenarios, potentially dangerous
scenarios, that you could then run in simulators, for example, or in the
nuclears, you can run these things as simulators and give operators
experience with them and see how they handle the situation.
DR. APOSTOLAKIS: So this would help with operator training?
MR. FORESTER: I believe it would, yes, because there is a
very explicit process.
DR. BONACA: I think we have the fundamental elements of
root cause, for example, and so, that would help with that.
MR. SIEBER: I think it also helps in revising procedures,
because you have a confusing procedure, and it doesn't really give you
the -- but this technique helps you pinpoint --
DR. APOSTOLAKIS: This is an important point that I think
you should be making whenever you make presentations like this, because
the sole objective is to support the PRA, and I think a legitimate
question would be are you sure you can quantify that? Maybe you can't,
but if your objective is also to develop health of operator training and
other things, then, I think it's perfectly all right.
DR. BONACA: I think the value of this, you know, when I
looked at this stuff is that -- was in part, I mean, some of the issues
are based on the mindset that the operators have. Here, you have a
boundary where they believe they have the leeway not to follow
procedures; for example, the issue of not going to bleed and feed was
very debated in the eighties, because it seemed like an option was that
severe accidents, something, and if you look at the procedure, before
1988 or so, there was no procedure to do bleed and feed. I mean, simply
said, if you have a dry steam generator, do something. One thing you
could do was bleed and feed.
Well, then, leave it to the judgment of the operator to do
so. Well, today, you go into it. We learned that that was a mistake.
So we said the only thing you can do is bleed and feed, so do it, and
you put it in the procedure now, and they follow it now, but it took a
long time for the operators to convince them to go into it. I mean,
they didn't like that.
So I'm saying that in a model like this, it would help to
talk about some of the shortcomings.
MR. SIEBER: I'm pretty well convinced that even if you
didn't have a PRA, you could profit from looking at how --
DR. APOSTOLAKIS: And all I'm saying is that those
statements should have been made up front, because the review, then,
doesn't say what you are presenting, and I would agree. I agree, by the
way, that this is a very valuable result.
DR. SEALE: It's interesting, because the utility of this
method actually begins in terms of influencing procedures and so forth
before it gets terribly quantitative, and yet, it's the ultimate
objective, presumably, or let's say the most sophisticated use of it is
when it gets quantitative so that you can use it in the PRA, but it
strikes me that it might be when you talk about these other uses to
actually identify the fact that in its less quantitative form, it's
still useful --
MR. CUNNINGHAM: Yes.
DR. SEALE: -- in doing these other things, and that
supports the idea, then, that you can evolve to your ultimate objective,
but you have something that's useful before it ever becomes the final
product.
MR. CUNNINGHAM: That's very useful. We've talked about
that and those types of benefits, but we could make it clearer.
DR. APOSTOLAKIS: Okay; can we move on?
DR. KRESS: Before you take that slide off --
DR. APOSTOLAKIS: We have two members who have comments.
Dr. Kress?
DR. KRESS: The three sub-bullets under two, if I could
rephrase what I think they mean, you start out with some sort of set of
base case scenarios, and you look at that scenario and look at places
where the scenario could be described wrong, and it could go a different
way somehow all through it, so those are the vulnerabilities or place
where it could go differently than you think or might even go different.
And then, the abbreviations are the possible choices of these different
directions a scenario might go; it looks a whole lot to me like an
uncertainty analysis on scenarios, which I've never actually seen done.
So it looks to me like a continuum. I don't know how you would make
this a set of integers.
MR. CUNNINGHAM: We'll talk about that later.
DR. KRESS: You'll talk about that later?
MR. FORESTER: Yes.
DR. KRESS: Okay.
MR. CUNNINGHAM: We want to get to that later.
DR. KRESS: Okay, so I'll wait until you do.
DR. APOSTOLAKIS: Mr. Sieber?
MR. SIEBER: I have a question. When I read through this, I
had a sort of an understanding of what the performance shaping factors
were. It's all the things that go into the operator, like training, the
culture of the organization, mission of the crew, formal and informal
rules, et cetera. That to me makes this whole process unique to each
utility, because the performance shaping factors are specific to a unit.
And this stuff is not transferable from one plant to another; is that
correct?
MR. FORESTER: That is absolutely correct.
MR. CUNNINGHAM: The process would be transferable but not
the results. That is correct.
MR. SIEBER: So you just couldn't take some catalog of all
of these potential possibilities for error and move them into your PRA,
and anything that had any relevance to anything --
MR. CUNNINGHAM: The potentials and the experience base are
useful inputs, but they are not substitutes for the analysis of an
individual plant.
MR. SIEBER: Well, when you're doing, then, a retrospective
analysis, you have to do it with the crew who was actually on the shift,
and you will reach a conclusion based on that crew, not necessarily that
plant; certainly not some other plant; is that correct?
MR. KOLACZKOWSKI: That would be the best track, correct.
MR. SIEBER: Thank you.
MR. FORESTER: So, sort of the next critical step after the
issue has been defined and the scope of the analysis is laid out is to
identify the base case scenario. So we've got to go into a little bit
more detail about exactly what we mean by base case scenario.
Usually, the base case scenario is going to be a combination
of the expectations of the operators as to how the scenario should play
out given a particular initiating event.
DR. APOSTOLAKIS: So these are key words. You're analyzing
response to something that has happened.
MR. FORESTER: Yes.
DR. APOSTOLAKIS: You have a nice description in chapter 10
of the various places where human errors may occur. Essentially,
they're also saying there that we recognize that the crew may create an
initiating event, but that's not really the main purpose of ATHEANA.
MR. FORESTER: Right; that's -- yes, the crew could
certainly create an initiating event, but they still have to respond to
it once they create it.
DR. APOSTOLAKIS: Right; so, the understanding is what an
event three, now, in the traditional sense, and the operators have to do
something.
MR. FORESTER: Right.
DR. APOSTOLAKIS: Okay.
MR. FORESTER: Okay; so, we're looking at that kind of
scenario, and it is the expectations for operators and trainers as to
how that scenario should evolve, what, sort of, their expectations are,
combined with some sort of reference analysis. Again, that could be
some sort of detailed engineering analysis of how this scenario expected
to proceed, and again, that could be something from the FSAR.
DR. KRESS: Would the structure of ATHEANA allow you to do
essentially what George says it doesn't do, and that is go into how an
initiating event is created in the first place, if it's created by an
operator acting of some kind?
MR. FORESTER: Well, certainly, we could --
DR. KRESS: Because you're starting out with normal
operating conditions.
MR. FORESTER: Right; well, in terms of what the process
does right now, it doesn't really matter whether the initiating event
was caused by an operator or someone working out in the plant or some
sort of hardware failure.
DR. KRESS: I know, but I was trying to extend it to where
we could do some control over initiating events by looking at the --
MR. FORESTER: Well, we didn't explicitly consider that, but
certainly, you could, you know, begin to examine activities that take
place in the plant and sort of map out how those things could occur and
then sort of use the process to identify potential problems with those
processes that take place in the plant that could cause an initiating
event, so it certainly could be generalized in that way.
DR. SEALE: That's an interesting point, because we always
worry about completeness of the PRA, and this is another way to cut into
the question of what are the possible scenarios that can be initiated
and do my intervention mechanisms, cross-cut those scenarios to give me
relief.
DR. KRESS: Well, my concern was initiating event
frequencies are kind of standardized across the industry, and they're
not plant specific. They probably ought to be.
DR. APOSTOLAKIS: I think this operator-induced initiate is
more important for low-power and shutdown point.
DR. KRESS: Yes, that's where I had -- that's what I was
thinking of.
DR. APOSTOLAKIS: But anyway, if they do a good job here,
that's a major advance, so let's not --
DR. KRESS: Let's don't push it yet.
MR. KOLACZKOWSKI: I was just going to comment that, for
instance, if you could have as the base case scenario how an operator
normally does a surveillance procedure, and then, you could look at the
vulnerabilities associated with that in terms of how well is he trained?
How well is the procedure written? Et cetera. And then, the deviations
would be how could the surveillance be carried out slightly different,
such that the end result is he causes a plant trip, so we still think
the process could apply. It is true that in the examples right now
provided in the NUREG, we don't have such an example, but we don't see
why the process would not work for that as well.
DR. APOSTOLAKIS: Because in those cases, the fact that you
have different people doing different things is much more important, and
ATHEANA has not really focused on that. Dr. Hullnager observed that,
too. So, I mean, the principles would apply, but it would take much
more work, which brings me to the question: what is the consensus
operator model? Are you talking about everybody having the same mental
model of the plant?
MR. FORESTER: Yes; well, and the same sort of mental model
of how the scenario is going to evolve. So, if you ask a set of
operators and trainers how they would expect a particular scenario to
evolve in their plant, you would get some sort of consensus. We try and
derive -- the analysts would try to derive what that consensus was.
DR. APOSTOLAKIS: Now, again, one of the criticisms of the
peer reviewers was that you really did not consider explicitly the fact
that you have more than one operator, that you sort of lumped everybody
together as though they were one entity. So in some instances, you go
beyond that, and you ask yourselves do they think, do they have the same
mental model of a facility, but the so-called social elements or factors
that may affect the function of the group are not really explicitly
stated; is that correct?
MR. FORESTER: It is in some ways, in the sense that when
you look at a crew perform, you can identify characteristics of how
crews tend to perform at plants.
DR. SEALE: You can find the alpha mayo, huh?
DR. APOSTOLAKIS: By the way, John, you don't have have to
have done everything.
MR. FORESTER: And I was going to say, we have not
explicitly considered --
DR. APOSTOLAKIS: Okay; good; let's go on.
MR. FORESTER: -- the two dynamics, okay?
[Laughter.]
MR. FORESTER: But it's not totally out of it is what I'm --
the point I was --
DR. APOSTOLAKIS: I agree.
MR. FORESTER: Okay.
DR. APOSTOLAKIS: Because you're talking about consensus
over the model.
MR. FORESTER: That is correct.
DR. APOSTOLAKIS: So it's not totally --
MR. FORESTER: Right.
DR. KRESS: I am still interested in the consensus operator
model. Excuse me for talking at the table but --
MR. KOLACZKOWSKI: That's okay; we understand why.
DR. KRESS: But, you know, I envision you've got two or
three sets of operators, so you have maybe -- I don't know -- 10 people
you're dealing with, and they each have some notion of how a given
scenario might progress. My question is really, do you have a technique
for combining different opinions on how things progress into a consensus
model? Do you have some sort of a process or technique for doing that
that you can defend or an interim entropy process or something?
MR. FORESTER: We don't have an explicit process for that.
I think the analysts were going to base their development of the base
case scenario on what they understand from what the operators are
saying; from what trainers are saying; what they see done in the
simulators when they run this kind of initiator in the simulator, how
does it evolve? Again, you have reference case.
DR. KRESS: It's a judgment.
MR. FORESTER: It is a judgment.
DR. KRESS: Of who is putting together --
MR. FORESTER: Yes, it is.
DR. KRESS: -- your model.
MR. FORESTER: Yes.
DR. KRESS: Okay.
MR. FORESTER: Okay; well, there's what we see as the
critical characteristics of the base case scenario, the ideal base case
scenario is going to be well-defined operationally; the procedures
explicitly address it; those procedures are in line with the consensus
operator model; well-defined physics; well-documented. It's not
conservative, and it's realistic. Again, we're striving for a realistic
description of expected plant behavior, so that then, we can try and
identify deviations from those expectations.
One thing I do want to note, that part of what is done
usually in developing the base case scenario is to develop parameter
plots, so that if a given initiating event occurs, we try and map out
how the different parameters are going to be behaving, but the
expectations of the parameter behavior will be over the length of the
scenario, because that's what the operators deal with. They have
parameters; they have plant characteristics that they're responding to.
So we try and represent that with the base case. And not every issue
allows that, but in general, that's the approach we want to take.
DR. POWERS: You have based those ideal scenarios on the
FSAR, you have you looked at how they deviate from the FSAR?
MR. FORESTER: That's right; okay; the next step, then, is
to see if we can identify potential operational vulnerabilities in the
base case. The idea is to try and find sort of areas in the base case
where things are not perfect, and there could be some potential for
human error to develop. We look for biases in operator expectations, so
if operators have particular biases, maybe they train a particular way a
lot, and they've been doing that particular training a lot; the idea is
to look at and try to identify what it is they expect and see if those
expectations could possibly get them to trouble if the scenario changed
in some ways, if things didn't evolve exactly like they expect them to.
DR. APOSTOLAKIS: So you are not really trying to model
situations like the Brown's Ferry, where they did something that was not
expected of them with the control rod drive pumps to cool the core? You
are looking for things that they can do wrong, but you're not looking
for things that they can do right to create -- because I don't know that
that was -- what was the base case scenario in that case, and what it is
it that made them take this action that would raise the core?
MR. FORESTER: I'm not sure I understand the -- no, no, yes,
the Brown's Ferry fire scenario.
DR. APOSTOLAKIS: Yes, the fire. They were very creative
using an alternative source of water.
MR. KOLACZKOWSKI: George, like PRA, this is basically, yes,
we're trying to learn from things that the operator might do wrong.
This is in PRA; we try to -- we treat things in failure space and then
try to learn from that. But we certainly consider the things that the
operator could do right, and particularly when we get to the recovery
step, which we'll get to in the process, in the case of the Brown's
Ferry fire, one of the things that the -- if we had now -- were doing an
ATHEANA analysis, if you will, of that event, a retrospective analysis,
one of the things you would recognize is that there was still a way out,
and that was to use the CRD control system as an injection source, and
that would be a recognized part of the process.
But, yes, just like PRA, we are basically trying to find
ways that the scenario characteristics can be somewhat different from
the operator's expectations, such that the operator then makes a mistake
or, if you will, unsafe act, as we call it, unsafe in the context of the
scenario, and ends up making things worse as opposed to better, and
then, we hope to learn from that by then improving procedures or
training or whatever, based on what the analysis shows us the
vulnerabilities are. So --
DR. APOSTOLAKIS: The emphasis here is on unsafe acts.
MR. KOLACZKOWSKI: That's what I ended up trying to figure
out is what could be the unsafe acts? What could be the errors of
commission or omission? How might they come about, and then, what can
we learn from that to make things better in the future?
DR. SEALE: But it still would be useful to understand what
it takes to be a hero.
MR. KOLACZKOWSKI: I agree it's still part of the recovery.
DR. BONACA: In all of the power plants, that's what people
refer to as tribal knowledge, especially discussions of the operators in
the crews and among themselves: what would you do if this happens and
so on? That would demonstrate the ways to get there, and in some cases,
they lead you to success, like, for example, the example you made here,
they would proceduralize and yet, they succeeded.
In the other cases, I've noticed things that they have that
they were talking about that would never lead to success; for example,
the assumption that, you know, you dry your steam generator, and now,
you do something to put some water in it; well, it doesn't cool that
way. You've got to recover some levels before you can do that. So the
question I'm having is is there any time to -- or is there any
possibility? I guess you can incorporate the type of information into
this knowledge, right? You would look for it. Is there any extended
process to look for it that you would model with ATHEANA?
MR. CUNNINGHAM: I think we'll come back to that.
DR. BONACA: The reason that I mentioned it is that that is
-- you know, if you look at a lot of scenarios we have in accidents, it
has a lot of that stuff going on. As soon as you get out of your
procedures, it comes in, and people do what they believe that --
DR. APOSTOLAKIS: In other terms, this is called informal
culture.
MR. FORESTER: That's right, and we are taking steps to
address those things; we certainly do.
DR. KRESS: I'm sorry to be asking so many questions, but
I'm still trying to figure out exactly what you're doing. If I'm
looking at, say a design basis accident scenario, what I have before me
is a bunch of signals of things like temperatures, pressures, water
levels, maybe activity levels in the various parts of the plant as a
function of time. This is my description of the progression of events.
Now, when you say you're looking for deviations that might cause the
operator to do something different than what -- are you looking for
differences that might exist in those parameters? The temperature might
be this at this time, or the water level might be this?
MR. FORESTER: It might change at a faster rate than this.
DR. KRESS: It might change at a faster rate than you
expect. So, those are the indicators you are looking at.
MR. FORESTER: Exactly.
DR. KRESS: And you're looking at how those might possibly
be different from what he expects and what he might do based on this
difference.
MR. FORESTER: Right.
DR. KRESS: Okay; thank you.
MR. FORESTER: Okay; so, there are essentially several
different approaches for identifying the vulnerabilities is what we have
up there. Again, we want to look for vulnerabilities due to their
expectations. We also want to look at a time line or the timing of how
the scenario should evolve to see if there is any particular places in
there where time may be very short, so if the scenarios are a little bit
different than expected, then, there should be some potential for
problems there, again, focusing on the timing of events and how the
operators might respond to it.
We also then tried to identify operator action tendencies,
so this is based on what we call standardized responses to indications
of plant conditions. Generally, for PWRs and BWRs, you can look at
particular parameters or particular initiators, and there are operator
tendencies given these things. We try and examine places where those
tendencies could get them in trouble if things aren't exactly right.
And then, finally, there is a search for vulnerabilities
related to formal rules and emergency operating procedures. Again, if
the scenario evolves in a little bit different way, the timing is a
little bit different than they would expect, there is some chance that,
again, even though the procedures may be technically correct, there may
be some ambiguities at critical decision points. Again, we try and
identify where these vulnerabilities might be.
And once we've identified those vulnerabilities, we go to
the process of identifying potential deviation scenarios. And again, by
deviations, we're looking for reasonable plant conditions or behaviors
that set up unsafe actions by creating mismatches. So again, we're
looking for deviations that might capitalize on those vulnerabilities,
and we're looking for physical deviations, okay, actual changes in the
plant that could cause the parameters to behave in unusual ways or not
as they expect, at least.
In this step of the process, we're also developing what we
call the error-forcing context. We're going to identify what the plant
conditions are. We want to look at how those plant conditions may
trigger or cause to become operable certain human error mechanisms that
could lead them to take unsafe actions and also begin to identify
performance shaping factors like the human-machine interface, recent
kinds of training they had that could have created biases that could
lead them, again, to take an unsafe action. So part of the deviation
analysis is to begin to identify what we call the error-forcing context,
and ATHEANA has search schemes to guide the analysts to find these real
deviations in plant behavior, and again, we are trying to focus on
realistic representations.
Part of the deviation analysis does involve, also, again,
developing parameter plots that try and represent what it is the
operators are going to be seeing and what is going to be different about
the way this scenario would evolve, the deviation scenario would evolve
relative to what they would. So these four basic search schemes that we
use to identify potential characteristics for a deviation scenario,
there are similarities between these searches; there is overlap. They
use similar tools and resources. There are a lot of tables and
information in the document to guide this process, but in general, we
recommend that each step is done sequentially, and by doing that, some
new information could come out of each step.
DR. APOSTOLAKIS: John, this is a fairly elaborate process,
and shouldn't there be a screening process before this to decide which
possibly human actions deserve this treatment? This is too much. Am I
supposed to do it at every node of the event three? If I look at an
event three, for example, and it has some point, you know, go to bleed
and feed, I know that's a major decision, major human action. I can see
how it deserves this full treatment, but there are so many other places
where the operators may do things here or there.
Surely, you don't expect the analysts to do this for every
possibility of human action. So shouldn't there be some sort of a
guideline as to when this full treatment must be applied and when other,
simpler schemes perhaps would be sufficient? Because as you know very
well, one of the criticisms of ATHEANA is its complexity. So some
guidelines before you go to the four search schemes, so right after, as
to which human actions deserve this treatment --
MR. FORESTER: Correct.
DR. APOSTOLAKIS: -- would be very helpful.
MR. FORESTER: Well, just a couple things. One is there is
an -- you know, if you identify a particular issue that you're concerned
with, then, you can identify what particular human failure events you
might be interested in, okay, or unsafe actions, so the issue may help
you resolve some of that in terms of what you would like to respond to.
If that's not the case, if you are dealing more with a full PRA, you're
trying to narrow down what it is you want to look at, then, we do
provide some general guidance in there for how to focus on what might be
important scenarios to initially focus your resources on.
DR. APOSTOLAKIS: But you're not going to talk about it.
MR. FORESTER: No, I hadn't planned on talking about that
explicitly. It's -- you know, I mean, you can say that, you know, it's
the usual kind of things, I guess, in terms of looking for -- trying to
prioritize things, you know, do you have some short time frame kinds of
scenarios? We have a set of characteristics; they're not coming to mind
right at this second, but a set of characteristics that were used to
prioritize those scenarios to focus on.
On the other hand, I think that the process itself, the
search for the deviation scenarios, you are reducing the problem,
because you're trying -- you're narrowing down to the problem kind of
scenarios. Okay; once you've identified, you know, an initiator, for
example, and maybe you're going to focus on several critical functions
that the operators have to achieve to respond to that initiator, then,
what the process does is it focuses the analyst in on the problem
scenario. So the process itself reduces what has to be dealt with.
We're not trying to deal with every possible scenario; we're trying to
deal with the scenarios that are going to cause the operators problems.
MR. KOLACZKOWSKI: Let me also add, George, though, I think
if you were going to apply this to an entire PRA, if your issue was I
want to redo the HRA and the PRA, I would say that no matter what HRA
method you used, that's a major undertaking.
DR. APOSTOLAKIS: Yes, but you are being criticized as
producing something that only you can apply.
MR. KOLACZKOWSKI: I was going to say -- thanks, Ann -- I
think you'll see, as we go through some more of the presentation and
show you the example, the method now has become much more -- excuse me,
methodical, and the old method that you saw in Seattle, it has changed
actually quite a bit from that method now. It's far quicker to use as
long as you don't want to get caught up in all of the little minute
documentation. You can actually do an entire scenario, set of
sequences, probably in a matter of hours to a day kind of thing.
MR. FORESTER: Once you've done a little bit of front end
work on this.
[Laughter.]
MR. FORESTER: So again, though, I do think the process
itself -- you're looking for the deviation scenarios; I think that
narrows the problem solving. Is that -- you know, the prioritizations
-- okay; okay, we have four basic searches. The first search involves
using HAZOP guide words to try and discover troublesome ways that the
scenario may differ from the base case. So again, we try and use these
kinds of words to ask questions like, well, is there any way the
scenario might move quicker than we expect it to or faster? Could it
move slower? Could it be more, in some sense, than what they expect,
given a particular initiator? For example, maybe given one initiator,
you also have a loss of instrument error. So now, it's more than it
was.
Another example might be in one of our examples in the
document is we're a small loca, close to a small loca, but it's actually
more than a small loca; yet, it's not really a large loca either. So,
again, we begin to look -- one way is to use these HAZOP guide words
simply as ways to investigate, you know, potential ways that the
scenario might deviate from what is expected, and the -- we're
interested in the behavior of the parameters, once again: are the
parameters moving faster than we expected in things like that? So
that's one way we do the search.
Another search scheme is then to identify that given the
vulnerabilities we already identified, maybe with procedures and
informal rules, are there particular ways that the scenario might behave
that could capitalize on those vulnerabilities? Should the timing
change in some way to make the procedures a little bit ambiguous in some
ways? That type of thing.
Third, we look for deviations that might be caused by subtle
failures in support systems, so this is sort of the way the event occurs
and the way something else happens might cause the scenario to behave a
little bit differently. They might not be aware that there is a problem
with the support system. So again, a subtle failure there could cause
them problems in terms of identifying what's happening.
DR. APOSTOLAKIS: Are you also identifying deviations that
may be created by the operators themselves, by slips?
MR. FORESTER: Yes, I don't see why we couldn't do that. I
mean, to arbitrarily examine what kinds of slips are possible at this
point in time, I'm not sure we've done that explicitly, but that's
certainly an option in terms of doing the deviation search.
DR. APOSTOLAKIS: Because it has happened.
MR. FORESTER: That's going to get pretty complex but --
DR. APOSTOLAKIS: It has happened.
MR. FORESTER: It has happened; that's true.
DR. APOSTOLAKIS: That isolated systems simply by their own
problem, but then, it takes about a half an hour to recover.
MR. FORESTER: Yes; I guess, you know, if we found some
vulnerabilities or we found some inclinations or some situations where
they might be focusing on particular parts of the control room or
something or on the panel, part of what we do examine are performance
shaping factors like the human-machine interface that could contribute
to the potential for an unsafe action, and in examining those things, we
would determine that there is some poor labeling or something that
creates the potential for a slip, that would certainly be figured into
the analysis.
DR. APOSTOLAKIS: So it could be, but it's not right now.
MR. FORESTER: No; I guess I shouldn't have said it that
way. I think it is. As I'm saying, once you've identified potential
deviations, part of the process is involved in looking at the
human-machine interface; looking at other performance shaping factors
that could contribute to the potential of the unsafe action. So, and
that is part of the process. That is explicitly part of the process, to
examine those things. So you might, then, identify, you know, it would
take someone knowledgeable about the way the control room panels and so
forth should be designed to maybe identify those problems, but
presumably, you'll have a human factors person on the team.
DR. KRESS: I'd like to go back to my question about the
continuous nature of deviations. Let's say you have a base case
scenario, and you've identified in there a place along the time line
that's a vulnerability and that the operator might do something, and
then, when he does that something, it places you in another scenario
that's different than your base case.
MR. FORESTER: Right.
DR. KRESS: And then, there are things going on after that,
and there may be different vulnerabilities in that line than there were
in the base case.
MR. FORESTER: That's true.
DR. KRESS: And there's an infinite number of these. I just
wonder how you deal with that kind of --
MR. FORESTER: Well, we try and deal with it during the
recovery analysis, when we move to quantification, when we try and
determine whether -- what the likelihood of the unsafe act might be.
Once they've taken that action, we then try and look at what kind of
cues would they get, what kind of feedback would they get about the
impact that that action has had on the plant; you know, what other
things; how much time would be available; what other input could they
receive in order to try and recover that action.
DR. KRESS: So you did tend to follow the new scenario out
--
MR. FORESTER: Right.
DR. KRESS: -- to see what he might be doing.
MR. FORESTER: Exactly.
Okay; and before I go to the last search scheme, I'd like to
go to the next slide. Actually, we sort of cover it on the next slide
anyway so --
DR. KRESS: When you say search, what I'm envisioning is a
person sitting down and looking at event trees and things and doing this
by hand. This is not automated.
MR. FORESTER: It's not automated at this point, no.
DR. KRESS: You're actually setting --
MR. FORESTER: It could be automated, yes, and we hope to be
able to automate it, provide a lot of support for the process.
DR. BONACA: You know, I had just a question. You know, it
took a number of years to develop the symptom-oriented procedures, and
they really went through a lot of steps from what you're describing
here. In fact, it was a time-consuming effort that lasted years, and
they had operators involved. Have you looked at them at all to try to
verify, for example, the process you are outlining here? Because they
did a lot of that work that could be useful.
DR. KRESS: It sounds very similar to that.
DR. BONACA: Yes; I mean, they have to go through so many
painstaking steps; you know, is this action or recommendation in the
procedure confusing? I wonder if you had the opportunity --
MR. FORESTER: Well, part of our process involves doing flow
charts of the procedures, specifically investigate where the ambiguities
could occur. So we go through that process.
Now, in terms of have we actively tried to look at, you
know, validating the existing procedures? No, we haven't taken that
step. But I think the general consensus is is that there are -- the
procedures are not perfect; that things don't evolve exactly -- I mean,
there can be timing kinds of issues, and there can be combinations of
different kinds of parameters that can be confusing.
DR. BONACA: So I think that probably, they would exercise
at one point with one set of procedures is what rules would be a good
foundation for a code like this and furthermore would give you some
indication of the strengths you may have in the process here of
identifying things or only the key points that -- for example, the key
points that were then central to the discussions of an owners' group, so
that they can identify in this process what they were, and they actually
go through the same situations. So there is a lot that can be learned
to verify the adequacy of a tool like this.
MR. CUNNINGHAM: No, that's a good point. We'll follow up
with that somewhere along the line here.
MR. FORESTER: Okay; on the next slide, one thing I wanted
to emphasize that a major part of the first three searches while we're
looking for the expectations, and they're using the guide words to sort
of characterize the way the scenarios could develop, we're also trying
to evaluate what the effect of those deviations, what the effect of the
deviations could be on the operators. What we wanted to determine is
the way particular parameters behave or the way the scenario was
unfolding, could that trigger particular human error mechanisms that
could contribute to the likelihood of an unsafe act?
Also, are there other performance shaping factors that could
then, based on the characteristics of the scenario and potential human
error mechanisms, are there performance shaping factors that could also
contribute to that potential for an unsafe act? So we're doing that at
the same time we're developing the actual deviations, and one thing
we've done, which I'll talk about here somewhere, I think -- maybe not
-- is to try and tie particular characteristics of the scenario: are
the parameters changing faster than expected? Or are two of them
changing in different ways? And try to identify how the characteristics
of the scenario could elicit particular types of error mechanisms:
could it cause the operators to get into a place where they're in kind
of a tunnel vision kind of state? They're focused on the particular
aspects of the scenario, or do they have some kind of confirmation bias
developed, or based on their expectancies, they have, you know, a
frequency bias of some sort.
And then, we try and tie the behavior of the scenario, the
characteristics of the scenario, to potential error mechanisms and then
relate specific performance shaping factors to the potential for the
error. We have tried to provide some tables that make that process a
little easier, so we have -- essentially, we have made an effort to try
and tie those factors together much more explicitly.
So getting that process, then, the fourth scheme, the fourth
search, is to sort of do a reverse process. If once you identify
potential error types and tendencies or operator tendencies that could
cause the human failure events or unsafe facts of interest, then, you
simply use conjecture to try and ask are there any kind of deviations
that could make these things occur, that have the right characteristics
that could make these things occur. So it's sort of coming from the
other direction rather than starting with the physical characteristics;
you just kind of start with the human tendencies and see if there are
deviations that could cause that.
So with those four searches, we think we do a pretty good
job of identifying a lot of potential deviation kinds of
characteristics. Then, once that's --
DR. APOSTOLAKIS: Does everyone around the table understand
what an error mechanism and an error type is?
MR. FORESTER: Well, error types are fairly straightforward,
in the sense that it's just things that they could do that could lead to
the unsafe fact, like make a wrong response; skip a step in a procedure;
normal kinds of -- it's not a real sophisticated kind of concept there;
it's just things that they could do.
Error mechanisms, we're referring to, you know, essentially
things within the human, general processing, human information
processing characteristics, what their tendencies are, maybe some
processing heuristics that they might use; not everything is going to be
a very carefully analyzed, completely systematic kind of analysis.
They'll use bounded rationality, so people have sort of general
strategies for how they deal with situations. Now, most of the time,
those kinds of situations, those kinds of strategies can be very
effective, but in some situations, the characteristics of the scenario
that may, where those particular tests may apply, may lead to an error,
because they're misapplied. So that's how we're characterizing error
mechanisms.
DR. APOSTOLAKIS: Is the inclusion of error mechanisms in
the model what makes it, perhaps, a cognitive model? I've always
wondered about these things. Because you have included these error
mechanisms, you can claim that now, you have something from cognitive
psychology in there?
MR. FORESTER: Well, we have the error mechanisms. We also
have the information processing model, you know, the monitoring and
detection process; the situation assessment. The human error
mechanisms, to some extent, are tied to those particular stages of
processing, so, you know, we try and include all of that. In fact, the
use of the tables that address the error mechanisms is broken down by
situation assessment and monitoring.
DR. APOSTOLAKIS: We are going very slowly.
MR. FORESTER: Okay; well, I'm just about done.
Once you have identified all of the deviation
characteristics, basically, you've got to put them all together and
identify the ones that you think are going to be the most relevant,
okay?
We can look to that. And the final slide is, again, we just
want to emphasize that once we have identified what we consider a valid,
a viable deviation scenario that has a lot of potential to cause
problems for the operator, and we analyze that, we want to quantify the
potential for the human failure event to occur or the unsafe actions.
We can directly address the frequency of plant conditions; standard
systems analysis to calculate that. We can get the probability of
unsafe act and the probability of nonrecovery at the same time given the
plant conditions and the performance shaping factors.
We look at this thing in an integrated way, and we do want
to emphasize that, that we carry out the scenario all the way out to the
very end, in a sense, to the last moment, when they can have a chance to
do something. We consider everything that's going on, and then,
ideally, in my mind, in terms of quantifying that, we have the input of
operators and trainers. Once you -- for example, if you can set up the
scenario on a simulator, you can run a couple of crews through that.
You may not necessarily -- you're not using that to estimate the
probability, but what I like to look for is what it is the operators and
trainers, what they think will happen when their crews in the plant are
sent through that scenario.
If everyone pretty much agrees, oh, yes, you know, if that
happened like that, we would probably do the wrong thing, then, you have
a very strong error-forcing context, and quantification is simple. For
situations where that is not the case, where there are disagreements
about what happened or not or a lot of high expectation that the actual
unsafe actions would take place, then, we do not have a new or a special
approach for dealing with that problem for a couple of reasons: one,
none of the existing approaches are completely adequate as they are.
For one thing, we have no empirical basis from psychology to support
those kinds of quantifications, those kinds of estimates. It just
doesn't exist.
Nor do we have an adequate existing database of events that
we can base it on. So, getting that situation, our suggestion for now
is to try and use existing methods. However, I think there are some
things that we could do to improve our existing quantification process.
You know, part of what we're recommending is maybe use SLIM. Well, the
problem with SLIM, of course, is you don't have adequate anchors. It's
hard to determine what the anchors might be so you can actually use a
SLIM kind of process.
So one thing we'd like to investigate, I think, is how we
could identify some maybe anchor kinds of events; we could characterize
the events that we could pretty substantially determine what the
probability of that event was; characterize that event in some way, at
least maybe a couple of events on the continuum, so that then, when we
characterize events using the ATHEANA methodology, we would know roughly
where they fit along that continuum. Okay; so, that's one improvement
that we could make that we haven't made right now.
DR. BONACA: One question I have is that in your
presentation, you are discussing the operator, but there are operators
who operate. One thing is to talk about the operators in the control
room who have been trained on system-oriented procedures, and there,
it's pretty clear how you can define the problem. The problem is that
they're following a procedure to the letter, and then, if there is some
area where we have misdesigned the procedures, then, we mislead them,
and they may have to initiate something that they're not used to, and
that's all kind of stuff.
Life is pretty clear that in the operators in the plant,
they follow procedures to do maintenance, for example, it seems to me
that the way you would train those kinds of operators would be very
different from the ones in the control room, because there, they have
their options on the procedures, on how you use them and so on and so
forth. Also, the operators are at the mercy of other operators doing
other things with other systems. I think even if you talked about how
they would --
DR. APOSTOLAKIS: They haven't done that.
MR. FORESTER: No.
DR. BONACA: So when you're talking about operators, you're
talking about the ones --
DR. APOSTOLAKIS: A single entity. A single entity.
DR. BONACA: Yes.
DR. APOSTOLAKIS: In the control room.
MR. FORESTER: In the control room, that is correct.
DR. APOSTOLAKIS: I have a few comments. This is the only
slide on quantification?
MR. FORESTER: Yes.
DR. APOSTOLAKIS: So I will give you a few comments.
MR. KOLACZKOWSKI: Except for the example.
MR. FORESTER: Yes, we do have the example.
DR. APOSTOLAKIS: Okay; on page 10-7, coming back to my
favorite theme, item two, the error-forcing context is so non-compelling
that there is no increased likelihood of the unsafe act. If you really
want the error-forcing context, the error-forcing context is so
non-compelling that there is no increased likelihood -- I really don't
understand your insistence on calling it forcing.
MR. FORESTER: Well I guess --
DR. APOSTOLAKIS: You don't have to comment.
MR. FORESTER: We've also been criticized for using the term
error at all, okay? But the point we want to make is operators are led
to take these unsafe actions.
DR. APOSTOLAKIS: Forcing -- and later on, you say that the
probability, even if it's very relevant, will be something like 0.5.
MR. FORESTER: Yes; I know Strater uses error-prone
conditions or error-prone situations, so there are other terms.
DR. APOSTOLAKIS: You saw here the HEART methodology. Have
you scrutinized it? I'll give you some things that bother me. On Table
10-1, there are generic task failure probabilities, so that first one is
totally unfamiliar; performed at speeds with no real idea of likely
consequences, and there is a distribution between 0.35 and 0.97.
Then, it says that in Table 10-2, HEART uses performance
shaping factors to modify these things, and the first 10-3 is
unfamiliarity. So now, I have a generic description of a totally
unfamiliar situation that I have to modify because I'm unfamiliar with
it, and the factor is 17. It's the highest on the table. So I don't
know what that means. Either I was unfamiliar to begin with, and
second, there is a distribution in Table 10-1. Am I supposed to
multiply everything by 17? What am I doing? Am I multiplying the 95th
percentile by 17? Am I multiplying the mean by 17?
MR. FORESTER: It's just the action. It's just the
probability for the action.
DR. APOSTOLAKIS: It's not explained.
MR. FORESTER: We didn't really claim to completely explain
HEART in there. We're trying to provide some guidance.
DR. APOSTOLAKIS: You need to scrutinize it, I think, a
little better.
MR. FORESTER: I think you're right, and a lot of the
categories are not always easily used. It's not a perfect method.
DR. APOSTOLAKIS: And then you say that one of the modifiers
is a need to unlearn a technique and apply another that requires the
application of an opposing philosophy. I'm at a loss to understand how
you make that decision, that somebody has to unlearn something and apply
something else.
And then, there is a modifying factor of five if there is a
mismatch between the perceived and the real risk. I don't know what
that means, risk. If I were you, I would throw this out of the window.
You don't have to take all these great stuff you presented in the first
18 view graphs and then present this thing. You should do your own work
here, in my view.
As I said earlier, I thought that the quantification part is
not at the same level of quality as the rest of the report.
MR. FORESTER: Agreed.
MR. KOLACZKOWSKI: Agreed.
DR. APOSTOLAKIS: You are throwing away a lot of the details
that you took pains to explain to us. There are no error mechanisms
here anywhere. And I fully agree, by the way, with what you said about
the difficulty and, you know, there has to be some sort of a judgment
here. There is no question about it, and this committee will be very
sympathetic to that, but not this kind of thing. And this is old,
right? The reference is from 1988, way before ATHEANA came into
existence.
The thing that is really startling is that it is not very
clear how the error-forcing context is to be used. They mention SLIM.
I thought I was going to see here an application of SLIM with the
problems that you mentioned. Everybody has those problems; where you
would remedy one of the difficulties or weaknesses of SLIM, namely,
which performance shaping factors one has to consider. And I think your
error-forcing context or whatever you call it in the future is ideal for
producing those. I mean, you have done such a detailed analysis. Now,
you can say, well, a rational application of SLIM would require perhaps
a set of independent ESFs or mutually exclusive -- I don't know what the
right term is -- and these are derived from the error-forcing context we
just defined in this systematic way, and no one will blame you for that,
because, I mean, if you've worked in this field for a month, you can
realize that the numbers will never be, you know, like failure rates,
where you can have data and all of that, and the anchors, I think you
pointed out, is an extremely important point, and perhaps you can do
something about it to give some idea.
But this guy who developed HEART had no heart.
[Laughter.]
DR. APOSTOLAKIS: His task is unfamiliar, and then, they
modified because I'm unfamiliar with the situation? I mean, what is
this? And a factor of 17, right? You increase the probabilities by
approximately 17.
[Laughter.]
MR. FORESTER: The only advantage to that method is this guy
did claim that a lot of these numbers were based on empirical data.
DR. APOSTOLAKIS: And you know very well --
MR. FORESTER: Yes, well, okay --
DR. APOSTOLAKIS: -- what that means.
[Laughter.]
DR. APOSTOLAKIS: Now, another thing -- so I'm very glad
that you are not willing to really defend to the end chapter 10.
MR. FORESTER: No.
DR. APOSTOLAKIS: It's probably something you wouldn't be
working on. Okay; I'm very happy to hear that, I must say, because I
was very surprised when I saw that.
Now, the -- actually, some discussion is really great. The
figures there, there is some type of figure 10-1 is repeated twice.
Well, that's okay. There was one other point that I wanted to make
which now escapes me -- oh, this -- all the information processing
paradigm is not here, right? You are not really using that.
MR. FORESTER: Well, we're using --
DR. APOSTOLAKIS: All of this stuff, I didn't see it playing
any role, at least the way it is now.
MR. FORESTER: It's not explicitly represented; you're
right.
DR. APOSTOLAKIS: The way it is now; okay.
MR. FORESTER: In our minds, it's represented.
DR. APOSTOLAKIS: Oh, I know that the mind is a much broader
term.
Okay; I'm very glad for that.
Okay; the dynamic element, and I believe Hullnagel commented
on that, too. We were doing in a different context some retrospective
analysis recently at MIT of two incidents. One was at Davis-Bessey; the
other was the Catawba. And what you find there is that there are some
times, critical times, when the operators have to make a lot of
decisions. There's no question about it. That's why you ask about the
training, and, I mean, you don't really want to attack each one with a
full-blown analysis.
MR. FORESTER: Right.
DR. APOSTOLAKIS: But in one of the incidents, I think it
was the Catawba, there were two critical points. One was 6 minutes into
the accident; the other 9 minutes. Where they had to make some critical
decisions, and the contexts were different, there was a dynamic
evolution. In other words, at 9 minutes, they had more information;
they were informed that something was going on, so now, they had to make
an additional decision. This specific element, the dynamic nature of
the EFC, is not something that I see here, and perhaps it's too much to
ask for at this stage of development, but it appears to be important,
unless I'm mistaken.
In other words, is the error-forcing context defined as a
deviation from what's expected? And for this sequence, it's once and
for all?
MS. RAMEY-SMITH: No.
MR. CUNNINGHAM: No.
DR. APOSTOLAKIS: No, so you are following the evolution and
the information that is in the control room, and you may have to do this
maybe two or three times at two or three different --
MR. KOLACZKOWSKI: Exactly, George. We present this as a
very serial type of process. Your point is well taken. You really have
to iterate and iterate. I think in one of the examples that we have for
the loss of main feed water event, one of our deviation scenarios is X
minutes into the event, all of a sudden, the spray valve on the
pressurizer is called for, and it sticks.
DR. APOSTOLAKIS: Right.
MR. KOLACZKOWSKI: That changes the scenario; it changes the
operator's potential response, and that's carried through. So I think
we try to do that.
DR. APOSTOLAKIS: Okay.
MR. KOLACZKOWSKI: But clearly, we're still discretizing the
situation into pieces of time, yes.
DR. APOSTOLAKIS: Okay; good, so, the dynamic nature of that
is recognized; that's good.
Now, a recovery in this context, my impression is it means
recovering from errors that they have made, not recovery in the sense
that the average person or the plant person will use it to recover from
the incident. They are two different things, aren't they?
MR. KOLACZKOWSKI: Well, ultimately, we're worried about it.
The core damage is the situation we're worried about. We're ultimately
worried about recovering the scenario. So, as I said, it will go back
to a success path. But part of that recovery may be overcoming a
previous error or unsafe act that the operator has performed. So now,
something has to come in that changes his mind about what I did an hour
ago, I now recognize was a mistake, and now, I need to do this. So,
that could be part of the recovery, but ultimately, we're looking at
recoveries of the scenario, yes.
DR. APOSTOLAKIS: So both.
MR. FORESTER: Both.
DR. APOSTOLAKIS: Okay; well, fine; if the main thing was to
realize that you, yourselves felt that chapter 10 needed more work, so I
have no more questions.
DR. POWERS: But I may still.
DR. APOSTOLAKIS: I'm sorry; yes.
DR. POWERS: As you're willing to say that the system is
more complicated, how do we decide that it's better?
DR. APOSTOLAKIS: In my view, as I said earlier, the
emphasis on context, the extreme attention that they have paid to
context is a very good step forward. Other HRA analyses, they do some
of it but not -- the quantification part, I am not prepared to say that
it is better, but I am glad to see that they are not saying that either.
But I think this detailed analysis that you see, there are other
argumentations in scope, but that's expected.
I think it's a good step forward. It's a very good step
forward. If I look at the --
DR. POWERS: Maybe the question is just different. The
analysis is more complicated. Therefore, you wouldn't have to be
sparing in your application of it. How would we know when this
complicated system --
DR. APOSTOLAKIS: I asked them that question and
unfortunately, they got upset.
[Laughter.]
DR. POWERS: And when can I do something else, and what is
that something else that I should do?
DR. APOSTOLAKIS: I think the message is very clear,
gentlemen, that you have to come up with a good screening approach. You
can't apply this to every conceivable human action.
MR. CUNNINGHAM: That's right, and if we need to better
describe how to do that and take that on, we've already talked about
that as an issue in terms of next year's work or this year's work, that
sort of thing.
DR. APOSTOLAKIS: Speaking of years, Hullnagel points that
out, and I must say I'm a little disturbed myself. This project started
in 1992, 7 years. Do all the members feel that this is a reasonable
amount of time for the kind of work they see in front of them?
DR. KRESS: Well, we'd have to know whether this work was
continuously done and how many people --
DR. APOSTOLAKIS: Mr. Cunningham is here. He can explain
that to us. Were there any --
DR. POWERS: Well, come on, George. It's difficult, is it
not, to manage the NRC? And besides, on the performance that they want
--
DR. APOSTOLAKIS: No, but on the other hand, if I'm
presented with a piece of work, I mean, how much effort has been
expended on it is a factor in deciding whether the work is good or not.
DR. POWERS: It is? That stuns me. It certainly is not in
the thermal hydraulics community.
[Laughter.]
DR. APOSTOLAKIS: After such a powerful
argument --
[Laughter.]
DR. APOSTOLAKIS: I defer humbly to -- I withdraw my
question.
DR. SEALE: The thing is that the entropy is always
increasing, whether you do a damn thing about it or not.
[Laughter.]
DR. KRESS: Only in closed systems.
DR. BONACA: One thing that I'd like to -- I like the
process, et cetera. Still, it seems to me that the process doesn't
distinguish, for example, between the French situation and the American
situation. In the U.S., we have extremely detailed procedures that the
operators will live by, and literally 10 years were expended to put them
together, going through a process which was as thorough as this and
involved all kinds of people, from the operators to engineers to
everybody else. And it seems to me that -- I'm trying to understand if
I go to review a possible situation that develops in an accident under
the French plan, where, in fact, there isn't a structural procedure; I
understand how I would have used it.
In fact, I would use it to see if the operator was
discussing the elements and what kind of errors he will make. I would
make a hypothesis. But in the U.S., I would tend to say that applied in
a way to review the procedures that they followed to see what errors he
would make in the U.S. and to eliminate all of those elements that are
then focused purely on the many possible -- see what I'm trying to say?
I don't see any of the --
MR. KOLACZKOWSKI: Yes.
DR. POWERS: It seems to me that it would be that way
because of the tie to the DBAs. When you tie them to the DBAs, you've
only got one measure. You say, gee, I can use this just to make sure my
-- but I think that when you go into the severe accident space, and you
have multiple failures, this network of deviations, there is an infinite
net that they show, and it changes character.
DR. BONACA: It does. There are new procedures. It's
totally different. They're not at all looking at these DBAs. They're
looking at the air pressure, temperature condition, et cetera, is moving
in this direction; what are you going to do?
DR. BARTON: And you still have underlying error.
DR. POWERS: But still you have underlying a failure, and
when you go to multiple failures --
DR. BONACA: You do, and it makes an assumption that, you
know, you are going to a key procedure, because you have conditions that
will require your ECCS to come up, for example, so there are some entry
decisions you make, but then, especially for the EPGs, for BWRs, they're
extremely symptom-oriented. I mean, at some point, you forget where you
came from.
DR. POWERS: Even with the symptom-oriented, you do things
that apply to an area that ultimately get you to what's wrong.
DR. BONACA: I understand, but again, if it was a plant X,
and they would use this, the first thing I would do, I would go through
this process to understand where my procedures were invested billions of
dollars; you're correct. That's really what happened. I mean, if it
followed literally, then, it would be different in certain respects from
the application that we make for -- where I have no prescribed way, and
so, I may discover that that's why I led the operator in the situation
we are in.
Now, I don't know if this had to have a different
perspective when you apply it to our plants, which are going through
very structured procedures. It seems to me every scenario would be
still open if you review it in a way where everything is possible, and
yet, you're ignoring the existence of the framework, which is exactly
the pattern of the steps you're suggesting here.
MR. CUNNINGHAM: I guess my reaction is that I think we
would have to kick that around among the team as to implications of the
French style versus the American style and that sort of thing. I just
-- I don't think we've thought much about that.
DR. APOSTOLAKIS: It may require a designer approach.
We will recess for 12 minutes, until 10:35.
[Recess.]
DR. APOSTOLAKIS: We have about an hour and 5 minutes, so
you will decide how best you want to use it. It's yours.
MR. FORESTER: Okay; I think what we'd like to do is present
an example of application of the method to some fire scenarios. This is
part of another task that we have to apply ATHEANA to fire scenarios.
We want to sort of do a demonstration of the methodology for fire
applications, and Alan Kolaczkowski is going to present this.
DR. APOSTOLAKIS: We have this or we don't have this? We
don't have it. No, we don't have the report.
MS. RAMEY-SMITH: It hasn't been written.
MR. KOLACZKOWSKI: My name is Alan Kolaczkowski. I work for
Science Applications International Corporation. George, I'm one of the
new team members. I've only been around for about a year and a half so
--
DR. KRESS: You're saying we can't blame you.
MR. KOLACZKOWSKI: Blame? No, I guess you can't.
Okay; well, you've heard at least in the abstract now what
the methodology involves, and again, I think the important points is
that -- and I think George articulated this very well -- is that we're
really trying to look at the combination of how plant conditions can,
based on certain vulnerabilities either in the operator's knowledge
about how the scenario might proceed, weaknesses in the procedures,
whatever, how those two things may come together in a way that if the
scenario is somewhat different from, if you will, the base case scenario
that maybe the operator is prone to perform certain actions which would
be unsafe in light of the way the scenario is actually proceeding.
I want to demonstrate now, actually, the stepping through
the process that will make some of these things and some of these
abstract ideas perhaps a little bit more concrete, step through it by
actually showing you an example, and as John pointed out, what I want to
do is take you through a set of a couple of fire analyses that we've
done, and as Ann pointed out, this report is currently in process in
terms of being put together.
So, the first slide, what I'd like to point out here really
is focus primarily on the third bullet, unless you have questions on the
others, and that is if you look at current HRA methods and the extent
that they look at fire events, and certainly, this had to be done as
part of the IPEEE program by the licensees, et cetera, what you find is
that a lot of the current HRA methods look at the human reliability
portion of the issue pretty simplistically. Most of the IPEEEs, if you
look at them, what they've done is they've taken their human error
probabilities from the internal events, and they might put a factor of
five on it and say, well, the stress is probably higher because there's
a fire going on, and there's a bunch of smoke, et cetera, and that's
what we're going to use for our human error probabilities.
And there really is, for the most part, not a hard look at
what is the fire doing? How is the equipment responding? Might some of
those responses be erratic? How might that change the way the operator
responds during the scenario, et cetera? That kind of look at what the
human is doing is typically not looked at. It's treated pretty
simplistically, for the most part. And so, we thought that this was an
error that would be very fruitful for ATHEANA to look at in order to
look at the context of fires and how scenarios from fire initiators
might affect the way the operators will respond as the fire progresses
and so on and so forth. So that's kind of why we looked at this.
DR. APOSTOLAKIS: What is SISBO?
DR. POWERS: Self-induced station blackout.
DR. APOSTOLAKIS: What?
DR. POWERS: Self-induced station blackout.
MR. KOLACZKOWSKI: I'm going to describe that in the next
slide, I believe.
So we decided that this was a pretty fruitful area to look
at, and that's why we chose this one as a good example to present here
in front of the committee.
DR. POWERS: Do we have a good phenomenological
understanding of how the fire affects equipment and other things?
MR. KOLACZKOWSKI: I guess I don't know how to measure good.
I think we have some general ideas, but that's part of the problem is
that fires can affect equipment in many, many different ways, which can,
therefore, make scenarios be somewhat different than what we expect, and
it's these kinds of deviation scenarios that we're talking about.
MR. CUNNINGHAM: In parallel with our work on human
reliability analysis, we have a separate program that's looking at the
issue of modeling of fires in risk analyses.
DR. POWERS: They repeatedly tell me that they can't really
predict what -- that that's why their research needs to go on --
MR. CUNNINGHAM: Yes.
DR. POWERS: -- is because they don't know what kinds of
things will happen to equipment.
MR. CUNNINGHAM: That's right; both are viable subjects,
reasonable subjects for research.
DR. POWERS: And I have had the licensees in saying the
vicious and evil thing about the NRC staff, because they take too
conservative a position on fire-induced changes and things like that.
MR. CUNNINGHAM: Again, we have another program. Part of
the reason for picking the fire example was to try to bring some of
these -- bring the two programs a little closer together.
MR. KOLACZKOWSKI: The next slide, as you're going to see in
a moment, we picked two particular scenarios to look at, but first, you
have to understand a little bit what the plant design is like, at least
in a general sense, for dealing with fires and what this SISBO concept
is, because we did decide to look at a so-called SISBO plant.
This cartoon, if you will, is meant to at least show you
what the separation is typically like in a nuclear power plant for
dealing with fire, and then, as I said, I want to introduce the SISBO
concept. You can see here that if you look at the cabling equipment in
the plant and so on, typically, for Appendix R purposes and so on, in a
very simple, two-division kind of plant, you end up with separating the
cables in the various cable trays and having certain walls and rooms and
fire barriers, et cetera, between equipment such that all the division A
equipment is located somewhat separately and at least are protected from
a fire standpoint from division B equipment, and we see that displayed
in this cartoon.
Of course, plants have now a remote shutdown panel
associated with them. Usually, that remote shutdown panel has a limited
amount of instrumentation and controls associated with it for
controlling one of the divisions of equipment for shutting down the
plant safely should the operators have to leave the main control room,
which might be the case for fire in the control room area as well as, as
you'll see in a moment, if it's a SISBO plant, there are other reasons
why they may leave the main control room as well.
So anyway, we have this standard separation between the two
divisions, and that separation, to the extent possible, is maintained
all the way up through the cable spreading room, the relay room, the
main control room, where we have the various fire barriers and so on and
so forth. As I indicated, we have this remote shutdown panel, the idea
being that if we need to leave the main control room, we go down to the
remote shutdown panel as well as other local areas in the plant, and we
operate this -- what's called dedicated areas of equipment or division A
equipment, and typically, what's done is that there is a set of switches
there on the remote shutdown panel, and that's just shown as one switch
in this little cartoon, that are thrown such that we become now isolated
from the main control room so that shorts, hot shorts or other
electrical problems that might be propagating up through the main
control room won't come down to the remote shutdown panel.
And now, we hook in the remote shutdown panel directly with
the equipment out in the field, and then we safely shot down the plant
from there. What's unique about the SISBO idea is that some plants, in
order to respond to various requirements in Appendix R and other
fire-related requirements for dealing with potential hot shorts and so
on, have taken on this so-called self-induced station blackout approach,
in which basically, what happens is the plant, once the fire gets so
severe that they feel that they are losing control of the equipment
because of erratic behavior, potentially because of hot shorts,
whatever, they essentially de-energize all of the equipment in the
plant, and at the same time, energize only either the alternate area
equipment if the fire is in a dedicated area zone, or they would go down
to the remote shutdown panel and operate the dedicated area of equipment
if the fire is in an alternate equipment zone and then re-energize that
equipment off that diesel. And then, they operate just that particular
set of equipment to safely shut down the plant.
So essentially, they put the plant into a loss of power
situation and then re-energize either A-bus or B-bus and then use just
selected equipment off of that bus that they think is not being affected
by the fire. Of course, the advantage of that is that now, hot shorts
can't occur in the A equipment, let's say if that's where the fire is,
because you've got it all de-energized, and so, you won't have a
spurious opening of the PORV or something like that that could make the
scenario much worse. So that's kind of the concept behind the SISBO
idea.
Next slide. Now, for illustrating the ATHEANA process, what
we've done is we've reanalyzed two fire scenarios that have been
previously analyzed in an existing PRA. This just highlights what the
two fires are and what the potential effects of the fires are for this
particular plant. One is an oil fire in the auxiliary feed water system
pump B room. This is for their classification, a so-called alternate
fire area, and you can see that if the fire does become significant, the
effects are quite severe. Four out of four of the non-safety busses
become affected and would potentially have to be shut down. You also
potentially lose the division B 4160-volt safety bus. That's the safety
bus for the various safety loads.
Of course, you lose, obviously, pump B of auxiliary feed
water, and it turns out in this particular plant, because of where the
cabling is located, if you had a severe fire in this room, you would
also affect the ability to operate and control the turbine pump. This
is a three-pump system that has two motor pumps, A and B, as well as a
turbine pump. This fire would affect one of the motor pumps as well as
the turbine pump.
If this situation got this severe, the expectations,
according to the procedures, would be that you would leave the main
control room, and then, you would shut down using limited division A,
that is, dedicated equipment, from the remote shutdown panel, and there
is an EOP, so called FP-Y, that governs how this is actually
implemented.
The other fire is, as I indicated there, a fire concerning
certain safety busses, and it turns out these safety busses are located
in the same area, room, if you will, that the remote shutdown panel is
located. So this is a so-called dedicated area fire, and again, if this
fire got severe, such that the feeling was that the operators were
losing control of the plant, the expectations, per the EOP, would that
-- well, first of all, you would lose the division A busses and the
ability to use that diesel and its various loads, and the expectations
would be you would shut down using division B equipment or so-called
alternate equipment.
In this case, they would stay in the main control room to
operate that equipment, but they're still going to de-energize
everything and then only energize the B busses and then use the B
equipment. So you're still going into a self-induced loss of power
situation.
Lastly on this slide, I wanted to indicate what the current
PRA insights are about the human reliability performance in these two
fires. And if you look at what are the sort of dominant lessons learned
from the HRA analysis for this existing PRA, those are highlighted there
on the third slide, that there is a potential for a diagnosis error to
even enter the right EOP, either EOP-Y if it's an alternate area fire or
EOP-Z if it's a dedicated area fire, so notice that one of the things
they have to know is where is the fire in order to know which EOP to
enter.
And the reasons why the existing human reliability analysis
technique says that a diagnosis error might occur are indicated here:
either the operator would misread or miscommunicate the cues to enter
the procedure, or he might just plain skip the step and not enter the
procedure or might misinterpret the instruction regarding when to enter
the procedure. Those were highlighted in the PRA as possible reasons
for why he might make this diagnostic error.
The more dominant errors, however, in the HRA, if you
actually look at the quantified results: they claim that it's much more
likely the operators will make mistakes in actually implementing the
EOPs themselves, just because they're very complex and so on and so
forth. There are a lot of steps involved.
Most of the errors, they claim, will be as a result of
switch positioning errors or just because of the fact that they may omit
certain steps because they're in a high stress situation. So that's
kind of what you learn from the existing PRA if you look at the human
reliability analysis for these two fires.
DR. POWERS: The regulation is that they're required to be
able to shut this plant down, so you're going to look at carrying out
that requirement.
MR. KOLACZKOWSKI: That is correct; we don't look at the
errors associated with still safely shutting down, but look at it now
from an ATHEANA perspective and say that if we think about the context
of these fires a little more, what might we learn that might be new,
more lessons learned that we could apply to ways to make the operators
better-prepared for dealing with these fires than just simply, well,
they might skip the step. Well, what are we supposed to do about that?
I guess we could say increased training, maybe, but we want to see if
ATHEANA can provide some additional insights as to how the operator may
not bring the plant back to a safe condition.
Yes?
DR. APOSTOLAKIS: Who did the PRA you are referring to?
MR. KOLACZKOWSKI: I'm sorry?
DR. APOSTOLAKIS: The PRA, the existing PRA. Is that the
utility?
MR. KOLACZKOWSKI: It is a -- yes, it's an IPEEE from a
licensee.
DR. APOSTOLAKIS: Okay.
MR. KOLACZKOWSKI: Now, John indicated that one of the first
things we do after really defining the issue, which, in this case, is
how can we learn better how the operators might make mistake given these
two kinds of fires and, therefore, take from that lessons learned and
ways to improve operator performance given these kinds of fires, once
we're able to identify that issue, one of the first things we have to do
is try to understand how does an operator, how does he think these two
fires would normally proceed?
This is that defining the base case scenario step. This is
trying to come up with that collective operator mindset as to what his
expectations would be given that these fires actually occurred, and our
base case is essentially summarized in this and the next slide, and let
me just kind of quickly go through this, and then, if you have any
questions, we can proceed to those.
Of course, one of the first things that would eventually
occur most likely is once the fire has happened, let's assume for the
moment that it happens without a person being in the room at the
particular time, et cetera; it's going to start to affect some
equipment, et cetera, but one of the first things that will probably
occur is that we will eventually get a fire detection alarm. There are,
at this plant, multiple alarms for detecting smoke, et cetera, in these
rooms and so on, so we would expect that fairly early in the scenario
that one of the first indications would be this fire detection alarm.
The operators then enter what is called EOP FP-X upon a fire
detection alarm, which basically provides the initial things that they
do for dealing with once a fire has been detected in the plant. One of
the first steps in that procedure is they ask another operator out in
the plant to go and visually validate that there actually is a fire,
that this is not a spurious or false alarm, and the procedure almost
reads as though the intent is that they don't do too much more until
that validation comes back.
Let's assume they do get the validation. Then, the fire
brigade is then assembled. It's called on. And one of the things they
do is they unlock the doors to the suspected area to make sure that the
fire brigade is going to have fairly easy access to that area, et
cetera, and there's a general notification over the Gaitronic system
that there is a fire in the plant and those kinds of things.
Now, during this time, especially if the fire is not yet all
that severe, the plant is still running. It's just humming along,
running along fine, and, in fact, the main control room staff are
attempting to just maintain the plant online and under proper control
while the fire brigade is now getting assembled and getting ready to do
their thing.
We expect that as time proceeds, and let's say the fire
brigade is finally getting down there, perhaps entering the room, et
cetera, but if the fire is getting to the point where it's approaching
the severities that I talked about in the previous slides, then, we're
going to start seeing erratic operation of some of the
normally-operating equipment. Perhaps we're going to start seeing flow
acting erratically; maybe if you have current indications on certain
pumps, like the AFW pump, you might begin to see erratic indications of
the current or maybe voltages on certain busses, depending, again, on
which cables are affected and when that occurs.
DR. POWERS: Isn't it much more likely that the things that
are going to be affected are the instrumentation and not the core
itself?
MR. KOLACZKOWSKI: That is true, too. I mean, it depends
on, looking at in each individual room, how much control and power
cables there are versus how much instrumentation cables. Certainly, the
AFW pump is instrumented to some degree, but the flow instrument for
flow going to the steam generator might be in an entirely different
room, and it's unaffected at all.
So it's very, very plant-specific, obviously, as to what the
specific effects are, but we would generally say erratic operation of
equipment, and certainly, your point is well-taken, Dana, of some
indications may be possible. But the point is the plant isn't
necessarily going to trip right away, and in a lot of small fires, as we
know, the plant ran through the entire scenario just fine. They put the
fire out, and that's it.
Now let's assume for the --
DR. POWERS: There is nothing at this point to indicate to
trip this plant.
MR. KOLACZKOWSKI: I'm sorry?
DR. POWERS: There is nothing at this point --
MR. KOLACZKOWSKI: No, FP-X does not require them at this
point yet to trip the plant. And, in fact, they will try to maintain
plant operation per their procedure at this plant.
So we have potential erratic behavior of some of the normal
operating equipment, perhaps some of the indications. Notice that
certain standby equipment may also be affected; for instance, that
turbine pump, the turbine auxiliary feed water pump, and it may also,
maybe, have cables associated with that pump's control that are burning,
and yet, they will have no necessarily idea that that pump has been
affected, because they haven't asked it to try to work yet. They're
still running the plant; feed water plants are still on. They'd have no
idea that the AFW turbine pump has now become inoperative. They won't
know that until they try to use it.
So just recognize that there is some missing information
with their situation assessment as to how bad this fire is, okay? Now,
also during this time; let's assume the fire brigade is trying to do its
job. There is going to be some diversion of attention as well, because
there's going to be periodic communication between the fire brigade and
the main control room staff. One of the things they do is hand out
radios, et cetera, and there's going to be talking back and forth: how
are you coming? What's the situation?
Maybe the brigade is saying, well, we haven't entered the
room yet; there's an awful lot of smoke, et cetera, et cetera. There's
going to be some diversion of attention dealing with the fire brigade as
well as trying to just make sure that the plant is okay. That's part of
the overall situation.
Let's assume for the moment that the conditions get even
worse. Either the fire brigade is having trouble getting out the fire
or whatever. At some point, if enough erratic behavior is occurring,
and we're actually beginning to have a lot of difficulty in actually
controlling the plant, maintaining pressurizer level, maintaining feed
water flows, whatever, that's when the judgment occurs for the operators
to then enter either EOP-FP-Y if the fire is in an alternate zone or
EOP-FP-Z if the fire is in a dedicated zone, and at that point, one of
the first steps in that procedure is, yes, trip the plant, okay?
Secondly, then, what they do after that is they, in the
procedures, is they basically isolate the steam generators, and then,
they leave -- if they have to, if they're in EOP-FP-Y, they have to
actually leave the main control room, and then, they start the
de-energization process, and that's when they actually are pulling
fuses, pulling breakers out locally in the plant, et cetera, and
essentially putting the plant into a self-induced blackout.
Simultaneously, they are -- and they actually take the crew
and separate them up into about three or four different areas of the
plant, so you have to also recognize that the crew is no longer working
as a unit in one room anymore; they're now located in various areas of
the plant talking on radios. One guy is over pulling fuses in a DC
panel; another person is over pulling breakers in an AC bus, et cetera.
So they're acting now certainly still in communication but as separate
entities.
They de-energize the various buses in the plant, and then,
they bring on the appropriate bus, depending on whether the fire is in
an alternate or dedicated zone, and then begin to bring on manually the
equipment they're going to use to safely shut down the plant. Now, in
the base case scenario, even if the fire got this bad, the expectations
of the operator would be, okay, we enter the right EOP procedure; we go
through its implementing steps; we carry it out; we eventually
restabilize the plant. Sometime during this time, the fire eventually
gets extinguished, and the scenario is over.
So in a general sense, this would be sort of the
expectations, even if the fire got fairly severe, as to what the
operators' expectations would be as to how the scenario would proceed,
and that's going to be our starting point to then build deviations on
that scenario.
One of the things we'll also do early on in the process is
we try to focus on, well, what human failure event or events and what
particular unsafe acts are we really interested in analyzing for? And
this slide is meant to attempt to try to summarize really the specific
human failure event that we're looking at, which is really failure to
accomplish heat removal. Let's say we get to the point where they have
to trip the plant, and now, they have to bring it back into a
stabilized, cooled state, recognizing they may have to leave the main
control room and go through this de-energization process and so on, and
what if they fail to carry that out correctly for one reason or another?
Taking that overall human failure event and really breaking
it down into, as we have here, three separate unsafe acts that we're
really going to be trying to analyze and determine, if we can, the
probability of that occurring. UA-1 is really very much closely
associated with that diagnostic error I talked about in the original
PRA; that is, one unsafe act could be the failure to enter the right EOP
or wait too long to enter that EOP, to the point where, perhaps by that
point, so much equipment damage has occurred; maybe hot shorts have also
occurred that they have essentially lost all control of the plant and
the ability to even bring it back to a cooled and safe and stable safe.
DR. APOSTOLAKIS: What's too long? Who determines the
length of fire?
MR. KOLACZKOWSKI: For purposes of this illustration, we
haven't tried to necessarily answer that question, George. It would
obviously depend on the specific plant; how big the fire grows; how fast
the equipment gets affected. You know, you could do that by doing
various com burn runs for that room and so on and so forth. It would be
very plant specific. I mean, I could try to give you some general
ideas, I suppose, but we have not tried to address that specifically in
this illustration.
DR. APOSTOLAKIS: Okay; but in terms of the base case
scenario --
MR. KOLACZKOWSKI: Yes?
DR. APOSTOLAKIS: -- do you have an idea as to how much time
they have? I thought that was one of the premises of defining the base
scenario.
MR. SIEBER: It depends on how big the fire is.
DR. APOSTOLAKIS: Well, okay, but they have to have some
sort of an idea how quickly they have to do it.
MR. KOLACZKOWSKI: I agree that as part of the base case
scenario, you would describe for a specific plant how long do they think
it would take before this fire would get that large and so on, and
that's going to be a very plant-specific answer.
DR. APOSTOLAKIS: I see Jack is shaking his head here.
MR. SIEBER: I don't think you can do it.
DR. APOSTOLAKIS: So how will the operators act?
MR. SIEBER: You act as quickly as you can without making
any mistakes.
[Laughter.]
DR. POWERS: What's happening in reality is that you've got
something, the fire alarm or something. You've got some people doing
things. They're talking to you about what they're finding. In the mean
time, you're going to have instruments that are telling you something is
going on, and the urgency, well, it's urgent to get the fire out, but
it's not urgent to take the plant, to trip the plant until you get
something urgent. Who says that? It's the instrumentation board or the
people that are talking about it. They say the fire is very big, and we
can't get it out with the people we've got; you're going to trip the
plant.
DR. APOSTOLAKIS: And this is now on the order of minutes?
DR. POWERS: Minutes.
MR. KOLACZKOWSKI: It could be.
DR. POWERS: Yes; I know. I mean, some of us are more
incredulous than others, but maybe that's just an area that somebody is
going to have to work on. It's in the area of most extreme abuse, I
think; what's already a very laborious process.
DR. APOSTOLAKIS: I think that's related also to the problem
of screening at the beginning. In other words, you really have to try
to make this not to look like it's an open-ended process that only a few
select people can apply.
I have another question. I'm confused there by the second
paragraph.
MR. KOLACZKOWSKI: Okay; I was going to get to that, George.
DR. APOSTOLAKIS: I think we have to hurry.
MR. KOLACZKOWSKI: Okay; go ahead.
DR. APOSTOLAKIS: Triggered error mechanisms include no
entry to procedures. And then, it says tends to lead to unsafe acts,
including taking no action. I thought the mechanism was something
different. I agree with the last statement, but if they delay or they
take no action, that's an unsafe act. I just don't see how it is an
error mechanism.
MR. KOLACZKOWSKI: Yes, it looks like maybe that is
miscategorized and should be down as an error type.
DR. APOSTOLAKIS: Okay; so it shouldn't be classified as a
trigger mechanism.
MR. KOLACZKOWSKI: I think I would agree with you, George.
DR. APOSTOLAKIS: Okay; I think we've got the flavor of the
search.
MR. KOLACZKOWSKI: Okay.
DR. APOSTOLAKIS: Unless the members want to see two, three
-- do you want to continue on to the deviation scenario development now?
MR. KOLACZKOWSKI: That's fine; that's fine.
DR. APOSTOLAKIS: Number 30?
MR. KOLACZKOWSKI: That's fine.
So we go through various searches to try to come up with
credible ways a scenario could be different, such that they trigger
certain error mechanisms that we think will lead to the error types of
interests, okay? Now, we actually -- once we've gone through those
searches, and we have some idea of credible ways that the scenario might
deviate from the base that really sets up the potential for the unsafe
acts that we're interested in, we then summarize those characteristics
into a troublesome scenario or scenarios; it might be more than one,
okay?
In this particular case, based on what we learned on the
searches in this illustration, we selected the following time line of
events that would be somewhat different. Imagine, if you will, that the
fire detection for whatever reason was delayed, either because of
perhaps some of the fire detection equipment not working and/or the fire
develops very slowly, which is getting sort of to the next bullet but
progressively.
Also, let's say the fire brigade has trouble putting out the
fire, although perhaps it reports back to the main control room that it
is almost under control. Obviously, with the kinds of things that
that's going to do, it's going to delay the decision process; allow the
potential for more equipment to be damaged before, in fact, the
operational staff take action; and if they're getting reports back by
the fire brigade saying we've just about got it out, again, the feeling
is going to be one of almost relief and say well, we're just about out
of this thing.
Now, beyond the initial fire conditions, also some other
later deviations that we're going to include in this "deviation
scenario" is that suppose that the fire duration and progression is such
that it gets so severe that it actually has cross-divisional equipment
effects. Perhaps it lasts longer than two or three hours, and
eventually, fire barriers get defeated or whatever, and/or other good
equipment, that is, the equipment they're going to try to use to safely
shut down the plant, what if it fails to function, like the diesel
doesn't start? Those that we think are credible, realistic deviations
in the scenario that could make the scenario much more troublesome.
Next slide.
DR. APOSTOLAKIS: So where are you using the fact that they
may be reluctant to abandon the control room?
MR. KOLACZKOWSKI: Well, again, that's been recognized as
part of one of the vulnerabilities, and the fact that we have a scenario
now that is going to develop slowly, and also, they're going to be
getting good reports from the fire brigade, we're basically saying
that's going to strengthen that reluctance. They're going to be less
willing to leave the main control room given that's the situation,
because they think the fire is just about out, and they're not sure what
all the effects of the fire are, in fact, because it's progressed so
slowly.
DR. APOSTOLAKIS: So that's not part of the deviation
scenario?
MR. KOLACZKOWSKI: It is a reason why the deviation scenario
is what it is. We're saying that this kind of a scenario, as described,
is going to strengthen or increase the reluctance factor. The scenario
is not the PCF. The scenario is described in an equipment sense.
DR. APOSTOLAKIS: What's the PCF?
MR. KOLACZKOWSKI: I'm sorry; I said PCS; PSS. The scenario
is going to strengthen certain performance shaping factors. In one case
here, one of the performance shaping factors, one of the negative ones,
is this reluctance.
DR. APOSTOLAKIS: So if one asks now what is the error
forcing context --
MR. KOLACZKOWSKI: Yes.
DR. APOSTOLAKIS: How many do you have, and which ones are
they?
MR. KOLACZKOWSKI: Okay; in this case, I guess we would say
we're describing one overall context. What you have before you on this
deviation scenario slide, the previous slide, is essentially the plant
conditions part of it. The actual performance shaping factors, I don't
think I have a slide on that, but the performance shaping factors which
make up the other part of the context would be things like unfamiliarity
with such a situation; reluctance to want to deenergize the plant and/or
if necessary leave the main control room and so on and so on.
And so, you would then describe those performance shaping
factors, and then, together, if you say given those performance shaping
factors and this kind of a scenario, we think we have an overall context
which may lead to higher probabilities of not entering the procedure in
time or carrying it out incorrectly, et cetera, those three UAs that I
talked about.
DR. APOSTOLAKIS: I mean, I thought that the error forcing
context is central to all of this. So I sort of expected the view graph
that said this is it.
MR. KOLACZKOWSKI: Probably should have stressed the
performance shaping factors; you're right. We only presented this --
DR. APOSTOLAKIS: Is it the performance shaping factors or
the context? Or these are part of the context?
MR. KOLACZKOWSKI: Yes; if you go back to the framework,
you'll notice that the error forcing context box has in it two things:
the plant conditions --
DR. APOSTOLAKIS: Yes.
MR. KOLACZKOWSKI: -- and the operator performance shaping
factors, and what we're saying is suppose the plant conditions are as
I've described in this deviation scenario. That's going to trigger a
lot of those other vulnerabilities that we talked about in the previous
step, which really become the performance shaping factors; that is, he's
going to have a reluctance to want to deenergize the plant, et cetera,
et cetera.
DR. APOSTOLAKIS: So you have a number of error forcing
contexts by selecting from the deviation scenario development.
MR. KOLACZKOWSKI: Yes, you could; yes, you could.
DR. APOSTOLAKIS: I think that's a critical --
MR. KOLACZKOWSKI: You could potentially have numerous
contexts.
DR. APOSTOLAKIS: You need to emphasize it and say these are
the contexts we're identifying.
MR. KOLACZKOWSKI: Okay; okay, good point.
Okay; given now we think we have a scenario that will, if it
develops in the way that we described in the deviation scenario, we
think along with the performance shaping factors provides us a more
error-prone situation or error forcing context, as we call it. One of
the things that we also do before we really enter the quantification
stage is think about well, what if it really did get this bad? What are
the potential recoveries?
I guess just quickly, for the case where he doesn't enter
the EOP or enters it way too late, we've assumed that if things got that
bad, right now for this illustrative analysis, we're not allowing any
recovery in that situation, and by the way, that's very similar to what
was done in the existing PRA. The existing PRA said if things get that
bad that he never made the decision to even enter the EOP, he's not
going to get out of this thing if the fire continues. So we're sort of
in line with what the existing PRA was in that case.
If the fire grows, and it affects both the alternate and the
dedicated equipment, which was one of the aspects of our deviation
scenario possibilities, well, obviously, now, now, the question becomes
what's he going to do, given he's got alternate equipment burning as
well as dedicated equipment burning, and really, there is no procedural
guidance for that. He's supposed to enter one or the other case, not
both. So if the fire grows and affects both the equipment, or, if when
he gets to the so-called good equipment, that is, the equipment not
affected by the fire that randomly fails, that could occur because of --
this is getting to your point, George -- the operator could be making
those problems occur, not just that the equipment fails.
This is sort of the operator inducing an initiator; in this
case, this is the operator actually causing the reason why the equipment
doesn't work. Maybe he doesn't try to start it up in the right sequence
or something like that, and so, it doesn't work properly.
Now, we have allowed recovery for that in the analysis, and
I think maybe the best thing I ought to do is go to the event tree,
which is the next slide, that will show the interrelationship of the
recovery with these unsafe acts. This is obviously very simplistic, but
what it's meant to do is cover really the key points that we're worried
about in how the scenario could progress. Notice we have the fire at
the beginning. Suppose the operator does not timely enter into the
correct EOP? That was the one that we said we're not going to allow a
recovery for. That's unsafe act number one. If that occurs, we're
going to assume for event tree purposes that that goes to core damage,
like the existing PRA did.
But suppose it does enter the procedure, and suppose the
fire does not jump to separation barriers; that is, it still remains in
only the alternate area or only the dedicated area. And then,
additionally, if the good equipment that he then tries to operate works,
well, that's the way out. That's the okay scenario he's trying to get
to. But if there is a problem either with the equipment working or if
the fire, in fact, jumps over into -- let's say it starts in the
alternate area and jumps to the dedicated area, maybe because of an
Appendix R weakness, or maybe there's a fire door inadvertently left
open, something like that, so the fire could get into the AFW pump A
room, for instance, as well.
Then, the operator is going to have to try to deal with this
situation that he's got fire affecting both alternate and dedicated
equipment, or he has to deal with the fact that the good equipment has
randomly failed and is not working, and when allowing a recovery there,
he has to make a decision as to what sort of recovery action to take,
and then, obviously, he has to carry out that recovery action.
That recovery action would probably be something like, well,
let me go try to use the A equipment again, even though it's the
equipment that's burning, because the B diesel isn't starting, so I've
got to go try to use the A diesel. That's my only out at this point.
So in event tree space, this is sort of the relationship between the UAs
and the equipment and the recovery and how that's sort of all panning
out.
DR. APOSTOLAKIS: Isn't this similar to an operator action?
MR. KOLACZKOWSKI: I guess certainly from the concept
standpoint, yes; in terms of laying out the possible sequences, yes.
Next slide. George, I don't know if you want to get into
the details --
DR. APOSTOLAKIS: No.
MR. KOLACZKOWSKI: -- of the codification other than to say
that we used the existing PRA information to try to quantify, well,
what's the chance this set of plant conditions would actually occur this
way. And then, as we said, as far as actually coming up with the
probabilities of the unsafe acts, at this point, they're still largely
based on judgment and using other types of techniques like HEART to try
to get some idea of what those numbers ought to be.
DR. APOSTOLAKIS: Why don't you go on to the difference
between existing --
MR. KOLACZKOWSKI: Okay.
DR. APOSTOLAKIS: -- PRAs?
MR. KOLACZKOWSKI: So that takes me to the last slide in my
presentation, which is really what we want to stress more than the
quantitative numbers. As with PRA, the real value of doing PRA is what
you get out of doing the process. The numbers are fine, and they sort
of set some priorities, but we think the same is true of ATHEANA. And
from a qualitative aspect, what we've done here is compare the existing
PRA human performance observations and sort of what you learned out of
the existing HRA and what you might learn out of doing an ATHEANA type
of HRA on these same two fires, and these are meant just to compare the
types of fixes or lessons learned, if you will, out of the HRA analysis
that one might gain from the existing PRA versus the ATHEANA results,
and let me just generally characterize them as I think the existing PRA
gives you some sort of very high level ideas of some things that you
might fix, and they generally fit the category of well, let's just train
them more, or let's make this step bolder in the procedure so he won't
skip it.
I think in going through the ATHEANA process and really
understanding what the vulnerabilities are and how the scenario
differences might trigger those vulnerabilities to be more prominent, I
think you learn more specifics as to ways to improve the plant, either
from a procedural standpoint, a labelling standpoint, et cetera, and
what the specific needs are, such as like that first one up there on the
extreme upper right. Clearly, there is a need for a minimum and
definitive criteria for when to enter EOP-FP-Y or Z.
DR. BARTON: That may be almost impossible to come up with:
how many meters; out of whack by how many degrees? Some of that is
going to be real hard to put numbers on, numbers or definite criteria
for getting in there.
MR. KOLACZKOWSKI: Granted; I'm not saying that all of them
can be done or should be done, but these are the types of insights one
can gain out of doing an ATHEANA type of analysis out of this. Unless
you want to go through specific ones, that pretty much ends the
presentation. It's trying to be a practical illustration of how the
actual searches and everything work.
DR. POWERS: I guess I'm going back to the question of what
has been accomplished? Why do we feel it's necessary to go to such a
heroic effort on the human reliability analysis? And if we could
understand why we want to do that, maybe we could decide whether we've
accomplished what we set out to do.
MR. KOLACZKOWSKI: My short answer to that is go back to one
of the first slides we had this morning. If you look at real serious
accidents, they usually involve operators not quite understanding what
the situation was; certain tendencies, et cetera, are built into their
response mechanisms, and therefore, they made mistakes, and PRAs, quite
frankly, as good as they do to try to determine where the risks of
nuclear power plant accidents lie, et cetera, still do not deal very
well with possible errors of commission, places where operators might
take an action that, in fact, would be unsafe relative to the scenario.
So maybe we're missing some of where the real risk lies.
DR. POWERS: I think we see this kind of a problem,
especially when we look at severe accidents, pertaining to accidents
where the operators disappear. Something happens to them, because they
don't affect things very much.
And you get peculiar findings out of that, like we have
people swearing that the surge line is going to fail; the four steam
generators to fail or the vessel fails, because that's where -- the
operator has apparently taken a powder and gone someplace and don't try
to put any water into it, and despite what we saw at TMI, the surge line
fails, and so, accidents become benign that otherwise would be -- and
understanding the operator is going to take a powder, that will do
something that seems like a very valuable thing.
The question you have to ask is is this enough, or should we
do something much more?
[Laughter.]
MR. KOLACZKOWSKI: I don't know how to respond to that.
DR. POWERS: Well, putting it another way, I assume you can
figure out the inverse to that statement, because that's already too
much.
MR. CUNNINGHAM: Part of the reason we're coming out to talk
to the committee and other people is just to sort out, okay, what are
the next steps? We've taken a set of steps. We've made an investment
and made a decision to go down a particular route.
DR. POWERS: Well, could you work and research just maybe
operators might put water in and the surge line not fail first?
[Laughter.]
MR. KOLACZKOWSKI: We'll do that. We'll try to convince
them.
DR. POWERS: Try to convince them that TMI actually did
occur.
[Laughter.]
MR. CUNNINGHAM: People forget things.
DR. POWERS: But it is possible that it pours down under
pressure and not had the surge line fail.
MR. CUNNINGHAM: Yes.
DR. APOSTOLAKIS: Are you going to be here this afternoon?
MR. CUNNINGHAM: I don't know about most of us
but --
DR. APOSTOLAKIS: Until about 3:00?
MR. FORESTER: I'd have to change my flight.
MR. CUNNINGHAM: Some of us will be here.
DR. APOSTOLAKIS: Okay; I propose that we recess at this
time so that Tom and I can go to a meeting, and we will talk about the
conclusion, followup activities at 12:45.
MR. CUNNINGHAM: 12:45 is fine by us.
DR. THOMPSON: I only have two more slides.
MR. CUNNINGHAM: We just have two slides, George, if you can
just bear with us.
DR. APOSTOLAKIS: Yes, but I want to go around the table.
DR. POWERS: Unfortunately, he has an hour and a half of
questions.
DR. APOSTOLAKIS: Yes.
[Laughter.]
DR. APOSTOLAKIS: Is the staff requesting a letter?
MR. CUNNINGHAM: We are not requesting a letter, no.
DR. APOSTOLAKIS: Okay.
MR. CUNNINGHAM: If you would like to write one, that's
fine, but we are not requesting it.
DR. APOSTOLAKIS: Okay.
DR. POWERS: We could write one on surge line failures.
[Laughter.]
DR. APOSTOLAKIS: So let's reconvene at 12:45.
MR. CUNNINGHAM: 12:45.
[Whereupon, at 11:45 a.m., the meeting was recessed, to
reconvene at 12:43 p.m., this same day.]. A F T E R N O O N S E S S I O N
[12:43 p.m.]
DR. APOSTOLAKIS: Okay; we are back in session.
Mr. Cunningham is going to go over the conclusions,
Catherine, so then, perhaps, we can go around the table here and get the
members' views on two questions: the first one, do we need to write a
letter, given the error forcing context that the staff is not requesting
a letter.
[Laughter.]
DR. APOSTOLAKIS: And the second, what do you think, okay?
So the staff will have a record of what you think. So, who is speaking?
Catherine?
DR. THOMPSON: Okay; just real quickly, I want to go over
two slides: the conclusion slide, we talked about all of this in the
last couple of hours that we think ATHEANA provides a workable approach
that achieves realistic assessments of risk. We can get a lot of
insights into plant safety and performance and have fixes, if you will.
DR. POWERS: It boils down to a lot on what you call
workable. It looks to me like it's not a workable approach. If I try
to apply it unfettered, I have some limitation on where I'm going to
focus it, but it completely gets out of hand very quickly.
MR. CUNNINGHAM: That's also true of event tree and fault
tree analysis and lots of other parts of PRA. I think one of the issues
that was discussed this morning of how do we fetter it, if you will, or
keep it from becoming unfettered, and I think that's a legitimate issue
that we perhaps can talk to you about more.
DR. POWERS: Yes; you need something that says, okay, you
need something that's a nice progression, so that you can go from
zeroeth order, first order, second order and have everybody agree, yes,
this is a second order application.
MR. CUNNINGHAM: Yes, yes, and that, I think, again,
probably within the team, we have those types of things in our heads.
DR. POWERS: Yes.
MR. CUNNINGHAM: But it's not very constructive from the
outside world, yes.
DR. APOSTOLAKIS: The same goes to a straightforward.
MR. CUNNINGHAM: Of course; it's intuitively obvious,
perhaps, that it's straightforward or some such things.
DR. POWERS: I got the impression that you had a variety of
search processes that made it comprehensive; they may not have made it
straightforward but a comprehensive search process.
MR. CUNNINGHAM: Okay.
DR. THOMPSON: Some of the followup activities.
DR. APOSTOLAKIS: Wait a minute, now, Catherine, you were
too quick to change that.
DR. THOMPSON: Good try.
DR. APOSTOLAKIS: This comes back to the earlier comment
regarding objectives. I don't think your first bullet should refer to
risk. Your major contribution now is not risk assessment. You may have
laid the foundation; that's different. But right now, it seems to me
the insights that one gains by trying to identify the contexts and so on
is your major contribution, you know, and that can have a variety of
uses at the plant and so on.
So I wouldn't start out by saying that you have an approach
to achieve a realistic assessment of risk.
MR. CUNNINGHAM: Okay.
DR. APOSTOLAKIS: You don't yet.
MR. CUNNINGHAM: Okay.
DR. APOSTOLAKIS: I, in fact, would make it very clear that
there are two objectives here, if you agree, of course. One is this
qualitative analysis, which I think I view as been knocked down a little
bit and then the risk part, okay?
MR. CUNNINGHAM: Yes.
DR. APOSTOLAKIS: I think you should make it very clear,
because if I judge this on the basis of risk assessment, then I form a
certain opinion. If I judge it from the other perspective, the opinion
is very different.
MR. CUNNINGHAM: Okay; I'll note that.
DR. APOSTOLAKIS: Develops insights: I have associated over
the years the word insights with failed projects.
[Laughter.]
DR. APOSTOLAKIS: Whenever some project doesn't produce
anything --
[Laughter.]
DR. APOSTOLAKIS: -- you have useful insights.
[Laughter.]
DR. APOSTOLAKIS: So in my view, you should not use that
word, even though it may be true.
MR. CUNNINGHAM: Okay.
DR. APOSTOLAKIS: Supports resolution of regulatory and
industry issues; you didn't give us any evidence of that, but I take
your word for it.
MR. CUNNINGHAM: Okay.
DR. APOSTOLAKIS: Okay.
MR. CUNNINGHAM: So insights will be removed from the
lexicon.
[Laughter.]
MR. CUNNINGHAM: Along with forcing, I guess, is another one
we have to remove.
DR. APOSTOLAKIS: Yes; the thing about unsafe acts and human
failure events, I really don't understand the difference.
MR. CUNNINGHAM: Yes; that's one of the things I was
thinking about this morning in listening to this is again, within the
team, I think it's well understood what those different terms means.
But to the --
DR. APOSTOLAKIS: Yes.
MR. CUNNINGHAM: -- the general public, it's not going to be
real clear.
DR. APOSTOLAKIS: But if it's an unsafe act, it should be a
failure demand? That's why it's unsafe?
MR. CUNNINGHAM: I don't know.
DR. APOSTOLAKIS: From the words, from the words; it doesn't
follow. And you are saying in the text that they are expected to act
rationally. So why are you calling what they did -- anyway.
MR. CUNNINGHAM: Anyway, yes, we will try to do a better job
of mapping those things out.
DR. THOMPSON: Okay.
MR. CUNNINGHAM: Followup issues?
DR. THOMPSON: These are some activities that we'd like to
get in a little bit more. Some of them are already planned.
DR. POWERS: You don't have any my surge line up there.
DR. THOMPSON: Surge line?
[Laughter.]
MR. CUNNINGHAM: There was a typo. We meant to say surge
line.
[Laughter.]
DR. POWERS: What you do is you didn't get the steam
generator tube rupture problems.
DR. THOMPSON: Okay; we obviously are pretty much done with
the fire issue. We're now working on PTS issue with Mr. Woods and some
other members of the branch and helping him look at the human aspects of
that. We'd like to get into some of the digital INC area, see what that
could add to the human error when they start working along with digital
INC.
DR. UHRIG: Are you looking at that strictly from the
operations standpoint, or are you going to get back into the code
development aspect?
DR. SEALE: The software side.
DR. THOMPSON: Software; we haven't -- these are things that
possibly we could get into. This isn't really planned yet, digital INC
part. So I don't know how far we would get into that.
DR. APOSTOLAKIS: So when you say digital, what exactly do
you mean? I guess it's the same question. The development of the
software or the man-machine interaction?
DR. THOMPSON: I think the man-machine.
MR. CUNNINGHAM: We were thinking not so much the
development as it's being used in the facilities.
DR. THOMPSON: Right.
DR. UHRIG: The difference between an analog and a digital
system is relatively minor when it comes to the interface. It's the
guts that's different. Pushing the wrong button, it doesn't make any
difference whether it's digital or analog.
MR. CUNNINGHAM: Yes; again, this has been suggested as a
topic that what we're doing here might dovetail well with other things
that are going on in the office. It hasn't gone much further than that
at this point.
DR. POWERS: At what point do we get some sort of comparison
of the leading alternatives to ATHEANA for analyzing human fault so that
you get some sort of quantitative comparison of why ATHEANA is so much
better than the leading competitors?
MR. CUNNINGHAM: A quantitative comparison or --
DR. POWERS: Well, a transparent comparison. You tried some
things where you said here's what you get from ATHEANA, and here's what
you get from something else. Any other different? But it's hard for me
to go away from saying this saying ATHEANA is just infinitely better
than the existing PRA results. Quite the contrary; I'm feeling that the
things in the existing PRA must be pretty good.
DR. BARTON: A lot of them are very similar.
DR. POWERS: Yes, pretty similar.
MR. CUNNINGHAM: Okay; they are similar but --
DR. BARTON: The whole process may end up fixed it sooner to
the fix out of play, the methods I'm using now.
MR. CUNNINGHAM: What happens in the context of like the
fire example is you're identifying new scenarios as you go through the
trees that seem to have some credible probability. How, you know, what
the value or what the probabilities are that will be associated with
them is still something we're still exploring. We expect that we will
find scenarios that will have a substantial probability and will, you
know, lead to unsafe acts or core damage accidents or whatever. Again,
they go back to you look at the history of big accidents in industrial
facilities, and you see these types of things occurring, so we're trying
to match the event analysis with the real world, if you will. In a
sense, that's one of the key tests, I think, of how well this performs
is that do we seem to be capturing what shows up as important in serious
accidents?
There are a couple of things that aren't on this slide that
we've talked about this morning. We discussed for a good while the
issue of quantification, that that may be -- is that on there? I can't
read the thing; okay, improved quantification.
DR. APOSTOLAKIS: What is that?
MR. CUNNINGHAM: It's one of those bullets.
DR. APOSTOLAKIS: Full-scale HRA/PRA?
MR. CUNNINGHAM: No, the fourth one down, improved
quantification tools.
DR. APOSTOLAKIS: I would say in degrading quantification.
MR. CUNNINGHAM: I'm sorry? Okay; quantification tools
comes up as an issue.
DR. APOSTOLAKIS: Why does the NRC care about whether
ATHEANA applies to other industries?
MR. CUNNINGHAM: Because it gives us some confidence that
it's capturing the right types of human performance. As we've talked
about many times or several times this morning, big accidents and
complex technologies, we think, have a similar basis in human
performance or are exacerbated or caused by similar types of events.
Given that we don't have many big accidents in nuclear power plants, I
think it's important that we go out and --
DR. APOSTOLAKIS: Did we ever apply this to other industries
to gain the same kind of lessons? Let them use it.
MR. CUNNINGHAM: Again, it's not so much the --
DR. APOSTOLAKIS: In my years at the Nuclear Regulatory
Commission, I don't know how much effort you plan to --
MR. CUNNINGHAM: Well, part of it, it's not a big effort,
but it's also something where I think it's important to help establish
the credibility of the modeling we have.
DR. APOSTOLAKIS: Like among pilots or airliners?
MR. CUNNINGHAM: Yes, the aircraft industry, over the years,
we've had some conversations with NTSB and with NASA and places like
that. Again, it's complex industries where you have accidents and --
DR. APOSTOLAKIS: I think developing quantification tools
and the team aspects in NNR will keep you busy for another 7 years, so I
don't know about the other industries. Again, that's my personal
opinion.
MR. CUNNINGHAM: Well, you can take that in several ways.
One of them is do you consider those the highest priority issues on the
--
DR. APOSTOLAKIS: I find them the most difficult, the most
difficult, applying it to other industries.
MR. CUNNINGHAM: I don't think we'd disagree with you.
DR. APOSTOLAKIS: I mean, it makes sense to -- adds
credibility to say, yes, we did it in this context and it's --
MR. CUNNINGHAM: Yes.
DR. APOSTOLAKIS: But I wouldn't put too much effort into
it.
DR. SEALE: But the preferable thing would be to have
someone else use ATHEANA, and then --
DR. APOSTOLAKIS: Yes.
DR. SEALE: -- you could get them to act as an independent
reviewer of your work and vice versa.
MR. CUNNINGHAM: Sure.
DR. SEALE: That strikes me as a much more --
MR. CUNNINGHAM: In that context, maybe apply is the wrong
word but interact with other industries --
DR. SEALE: Yes.
MR. CUNNINGHAM: -- complex industries on the -- for the
credibility and the application of ATHEANA.
DR. APOSTOLAKIS: Well, you also have, it seems to me, a
nuclear HRA community. Why are the teams developing whatever processes
or whatever? Is it because they're not aware of ATHEANA yet?
MR. CUNNINGHAM: You're taking some of the next
presentation, which is on the international work that we're doing.
DR. APOSTOLAKIS: I'm not sure that we're going to have that
presentation.
MR. CUNNINGHAM: Okay.
DR. APOSTOLAKIS: I think we should conclude by discussing
what we've heard, unless you really feel that -- I mean, I look at it.
It's not just really useful.
MR. CUNNINGHAM: No, no, I'm sorry; there's a separate
presentation.
DR. APOSTOLAKIS: There is?
MR. CUNNINGHAM: Yes; remember this morning that we
discussed -- one of the first things on the agenda was the work we're
doing internationally. We put that off until after that.
DR. APOSTOLAKIS: How many view graphs do you have on that?
MR. CUNNINGHAM: It's about eight or something like that.
We can cover it in 5 or 10 minutes.
DR. APOSTOLAKIS: I think we should do that right now.
MR. CUNNINGHAM: Okay; it's up to you.
DR. POWERS: I would hope you would be able to tell me that
little -- the Halden program plays or could play in the ATHEANA
methodology.
MR. CUNNINGHAM: Do you want to go ahead and go to the
international?
DR. POWERS: Whenever it's appropriate.
DR. APOSTOLAKIS: It's up to you, Mark. I think we're done
with this.
MR. CUNNINGHAM: We're done with this; then, let's go ahead,
and we'll cover the international thing.
DR. APOSTOLAKIS: I want to reserve at least 5 minutes for
comments from the members.
MR. CUNNINGHAM: Okay.
DR. APOSTOLAKIS: Before we go on to the Sorenson
presentation.
MR. CUNNINGHAM: Okay.
Basically, as we've been doing this ATHEANA work and our
other HRA work, we've had two principal mechanisms for interacting
internationally with other developers and appliers of HRA methods. One
is through the CSNI principal working group five on PRA; in particular,
there was something called the task group 97-2, which is working on the
issue of errors of commission.
DR. APOSTOLAKIS: Who is our member?
MR. CUNNINGHAM: I'm sorry?
DR. APOSTOLAKIS: Who represents the NRC there, PWG-5?
MR. CUNNINGHAM: We have two or three different
interactions. Joe Murphy is the chairman of PWG-5; I'm the U.S.
representative on 5; the chair of the 97-2 task group was Ann
Ramey-Smith. We also have our COOPRA programs. One of the working
groups there was established to look at the impact of organizational
influences on risk.
DR. APOSTOLAKIS: Is that what the Spaniards are doing?
MR. CUNNINGHAM: Yes, that's where the Spanish come in.
It's the international cooperative PRA research program. It doesn't fit
the --
DR. APOSTOLAKIS: That is one of the Former Chairman
Jackson's initiative papers.
MR. CUNNINGHAM: Correct; she wanted to -- she wanted the
regulators to work more closely together, and there were a couple of
research groups established as part of that.
Anyway, okay, the PWG-5 task 97-2 had three general goals.
You want to look at insights, although perhaps that's no longer the
right word to use; develop perspectives on errors of commission to apply
some of the available methods which supposedly handle errors of
commission and for quantitative and non-quantitative, more qualitative
analysis of errors of commission and to look at what data would be
needed to support types of analysis.
DR. POWERS: Have any of the technical fields -- I can with
modest amount of effort, have you seen the database that -- is there
someplace that I would go to find data that are pertinent to human
reliability analysis?
MR. CUNNINGHAM: Do you want to answer that? I'm going to
have one of my colleagues come up and answer that a little more
explicitly. One of the people over here was shaking her head; I don't
know.
MS. RAMEY-SMITH: No, that's a short answer.
[Laughter.]
DR. APOSTOLAKIS: Would you identify yourself please?
DR. POWERS: Before she identifies herself as a major expert
in the field that I noticed last year our first exposure to ATHEANA was
on human reliability analysis, brand spanking new, put out by a book
publisher, and so I immediately acquired a copy of this book; read it
for an entire airplane flight from Albuquerque to Washington, D.C. and
found not one data point in the entire book. But there were 30-some
papers on various human reliability analyses but not one data point.
DR. SEALE: We still need to know who she is. For the
record, please?
MS. RAMEY-SMITH: Ann Ramey-Smith, NRC.
If I can recall, the question was is there a database that
you can turn to, and the short answer from our perspective of the kind
of analysis that -- and from the perspective that we think you should do
an analysis, which is within the context of what's going on in the plant
and performance shaping factors and so on, there is not a database that
exists that we can turn to and go -- and make inferences based on
statistical data.
The fact is that we've developed our own small database that
has operational data in it that we have analyzed. There are various and
sundry databases of various sorts. The question comes down, and one of
the questions that this PWG-5 is going to address is the fact that we
have a lot of databases, none of which may serve the needs of the
specific methods that people are trying to apply.
DR. UHRIG: Would there not be a lot of information
available through the LERs?
MS. RAMEY-SMITH: Oh, if that were true. Actually, there is
quite a lot of information available on the LERs. Unfortunately, it's
difficult oftentimes in those writeups to understand fully what the
context was, to understand why the operators did what they did and what
were the consequences and what were the timing and so on and so forth.
One concern that some of the HRA folks have is that possible changes to
the LER rule will even strip from the reports the little information
that it had before, so we're concerned about that.
The better source for information, actually, has been the
AIT reports and some very excellent reports that were previously done by
AEOD when they did studies of particular events that maybe didn't rise
to the level of AITs but were very in-depth analyses, and we were able
to make use of those, particularly early on when we were doing this
iterative evaluation of operating experience. It was quite helpful.
DR. POWERS: One of the issues that NRR is having to
struggle with is these criteria in what actions should be automated as
opposed to being manual. How long does it take somebody to diagnose a
situation and respond to it? And there are several that they have,
because they have some good guidelines; they just don't have any data.
MS. RAMEY-SMITH: I think this approach would be very
helpful for understanding -- what is it? -- B-17, the safety-related
operator actions. I think that the agency would be wise to evaluate
that issue within the context of PRA.
DR. APOSTOLAKIS: This looks to me like a benchmark
exercise. Is that what it is?
MR. CUNNINGHAM: No; the sense that I have is that someday,
we might be able to get to a benchmark exercise, but the principal
players weren't comfortable at this point in constraining the analysis
to that degree.
DR. APOSTOLAKIS: So, oh, yes, because you're saying they
apply to events of the --
MR. CUNNINGHAM: That's right; we have a variety of
different methods, and what we were doing was trying to see what these
methods were giving us, so we didn't try to constrain it to a particular
method or a particular event.
DR. APOSTOLAKIS: Okay; thank you.
MR. CUNNINGHAM: As you can see on page 4, we have a number
of different methods applied. ATHEANA was applied by the U.S. group,
the Japanese in people in the Netherlands; also different methods
applied such as MERMOS, SHARP. We have the Czech Republic spelled
correctly today, so that was an advancement over yesterday.
[Laughter.]
MR. CUNNINGHAM: And some other models that, as you can see,
we go back to the Borsele theory.
DR. APOSTOLAKIS: Is SHARP really a model?
Okay; let's go on.
MR. CUNNINGHAM: Okay; slides five and six are a number of
the conclusions that are coming out of the task 97-2. I'm not sure I
want to go into any of the details today, but you can see the types of
the issues that they're dealing with and what the report will look like.
The report has been by and large has been finished; the report of this
group has been finished. It's going to go before the full CSNI next
month, I believe, for approval for publication. So it's essentially --
this part is particularly -- is essentially done.
DR. APOSTOLAKIS: The words are a little bit important here.
The rational identification of errors of commission is difficult. What
do you mean by rational?
MS. RAMEY-SMITH: That was the word that was chosen in the
international community that everyone was comfortable with. But the way
you can think of it is it's as opposed to experientially, you know, so
that it's more predicting to sit down and to be able to identify errors
of commission a priori.
DR. APOSTOLAKIS: Do you mean perhaps systematic?
MS. RAMEY-SMITH: Yes, that could have -- I guess the point
is to be able to I guess systematically analyze it, you know, a priori
be able to identify an error of commission. Systematic is a perfectly
good word. This was just the word -- we used on this slide the words
that, in the international group that was working on this, they were
comfortable with.
DR. APOSTOLAKIS: And what is cognitive dissonance?
MS. RAMEY-SMITH: Okay; perhaps Dr. Thompson would like to
--
DR. APOSTOLAKIS: That was an international term?
MS. RAMEY-SMITH: No, cognitive dissonance is from the good
old field of psychology.
DR. APOSTOLAKIS: Oh, okay.
DR. BARTON: It's Greek.
DR. APOSTOLAKIS: What?
DR. BARTON: It's Greek.
[Laughter.]
DR. SEALE: Could I ask if this group of international
experts had all of these different approaches, presumably, they would
have a great deal of common interest in making certain things like LERs
helpful about what's there. Has anyone put together a sort of a
standard format for what it would take to get an LER that had the
information you needed in it be able to generate a database?
MR. CUNNINGHAM: Actually, one of the follow-on tasks of
this work is for the HRA people here to go back and try to lay out what
data do they need based on their experience with this type of thing. So
today, I don't think we have it, but I think over the next year or so,
CSNI PWG-5 is going to be undertaking an effort to put that in the
lifestyle.
DR. SEALE: It seems to me that should be something you
could go ahead on, and whatever happens, at least now, you'll be getting
information that's complete --
MR. CUNNINGHAM: Yes.
DR. SEALE: -- in some sense.
MR. CUNNINGHAM: Yes.
DR. APOSTOLAKIS: That would be a very useful result.
MR. CUNNINGHAM: And that's one of the things that PWG-5 is
going to undertake.
MR. SIEBER: Does that mean that every LER a plant puts out
here goes through the ATHEANA program?
DR. APOSTOLAKIS: No, no, no, no. The ATHEANA has developed
guidance about the LERs. The guys who write the LERs don't need to know
about ATHEANA.
MR. CUNNINGHAM: Okay.
DR. SEALE: Just what it takes to have all of that planning
data and things like that in it so that you've got a picture.
MR. CUNNINGHAM: Just two clarifications. One was this
isn't the ATHEANA guys; it's the -- this international group of HRA
people, so it's the MERMOS guys and all those guys are going to be doing
it. It's not an ATHEANA specific issue.
The second, I was talking about data needs in general. I
wasn't trying to suggest that all of the data needs that we had would
automatically translate into something at LER, a change in the LER
reporting requirements. I wasn't suggesting that.
DR. APOSTOLAKIS: There has been a continuing set of
discussions on human liability, and as I remember, former member Jay
Carroll was raising that issue every chance he had. How can you
restructure the LERs so that the information is useful to analysts?
Because the LERs were not designed -- they were designed for the PRA
phase, right? You don't need another review for that.
MR. CUNNINGHAM: The LERs have a particular role, and as
that role is defined even today, it's not going to provide a lot of the
detailed information. Now in parallel, though, with the development of
all of the LER generation, you have the NPO and NRC and industry work in
EPIX, which will be collecting information that is much more relevant to
PRA types of analyses. So I wouldn't so much focus on LERs as EPIX.
DR. APOSTOLAKIS: It would be nice to influence what those
guys are doing.
MR. CUNNINGHAM: Yes.
DR. APOSTOLAKIS: Okay; next.
MR. CUNNINGHAM: Okay; going on to slide seven on the COOPRA
working group on risk impact of organizational influences, basically,
we're trying to -- the goal of the working group is to identify the
relationships between measurable organizational variables and PRA
parameters so that you can bring the influence in and explicitly model
the influence in PRAs.
DR. APOSTOLAKIS: Next.
MR. CUNNINGHAM: Overall, I don't think I need to go into
the outcomes as much as -- I think it's understood as to what that is.
Right now, it's fairly early in the process. We're trying to get a
better understanding of what people are doing in this area. You alluded
to the Spanish work in this area. The Spanish are one of the key
contributors in here. How many countries are involved in this?
MS. RAMEY-SMITH: It's about six or seven.
MR. CUNNINGHAM: Okay; about six or seven countries; the UK,
France, Spain, Germany, did you say?
MS. RAMEY-SMITH: Yes, Germany.
MR. CUNNINGHAM: Argentina, Japan?
MS. RAMEY-SMITH: Japan.
MR. CUNNINGHAM: Japan. They're trying to work together on
this issue. Basically, again, this is fairly early in the work here.
There's going to be another meeting early next year to basically take
the next step forward in the COOPRA work. That's --
DR. APOSTOLAKIS: That's it?
MR. CUNNINGHAM: That's the short summary of the
international work.
DR. APOSTOLAKIS: Okay.
DR. POWERS: And so, the Halden program has no impact on
your --
MR. CUNNINGHAM: I'm sorry?
DR. POWERS: The Halden program has no impact on your --
MR. CUNNINGHAM: The Halden program has traditionally -- Jay
Persensky sitting back here knows far more about it than I -- but has
traditionally been oriented towards not so much human reliability
analysis for PRA but for other human factors issues. There has been
some ideas that Halden will become more involved in human reliability
analysis. That's at least, I guess, in the formative stages.
MR. PERSENSKY: Jay Persensky, Office of Research.
Halden has proposed for their next 3-year program, which
starts in November, the development of an HRA-related activity based
primarily on input from PWG-5, because a number of the people that have
been involved with the Halden human error analysis project also serve on
that or have served on that task force. The goal, as I understand it at
this point, is aimed more towards trying to take the recommendations
with regard to kinds of data and seeing whether or not they can play a
role in that. At this point, it is in the formative stage, but it's
looking more at that aspect of data since they do collect a lot of data,
at least simulator data in-house.
Now, whether it can be used or not is another question. And
that's what they're looking at at this point.
DR. POWERS: Is cross-cultural data any good? In other
words, if I collect data on the Swedish or Norwegian operators on a
Finnish plant, is that going to be any good for human error analysis,
modeling or for American operators on American plants?
MS. RAMEY-SMITH: It has the same context.
MR. CUNNINGHAM: When you say data, it depends. If you're
talking about probabilities, I don't know that any of the particular
probabilities will apply, because again, there's a strong context
influence. Can it provide some more qualitative insights? I suspect it
could but again --
DR. POWERS: Cognitive things? What does it tell you about
processing information, things like that? Are there big enough cultural
differences that it's not applicable? I would assume that Japanese data
would just be useless for us.
MR. CUNNINGHAM: I wasn't thinking of the Japanese, but
there may be some cultures where it would be of real questionable use
depending on the basic management and organization and how they do
things and whatever, it could be and not be very applicable.
DR. APOSTOLAKIS: Okay; all right, why don't we go quickly
around the table for the two questions: Should we write a letter, and
what is your overall opinion?
Mr. Barton?
DR. BARTON: Yes; I think we need to write a letter. But
let me tell you what my opinion is first --
DR. APOSTOLAKIS: Okay.
DR. BARTON: -- and maybe we can figure out if my opinion is
similar to others; maybe not. I fail to see the usefulness of this tool
for the work that's involved. Maybe I need to see some more examples.
I mean, the fire example doesn't prove to me that ATHEANA is much better
than existing processes I know when looking at EOPs and how I train
people and how people use procedures or react to plant transients.
I think that as I look at this process, I also see where a
lot of some of these actions depend on safety cultures, conservative
decision making, et cetera, et cetera, and those two tie into this to
understand more help and more safety culture and conservative decision
making also.
I think the tool -- I don't want to poo poo the tool, but I
think it's a lot of work, and I don't see that you get a lot of benefit
out of going through this process to really make it something that
people are going to have to use in their sites unless this is a
voluntary thing. I don't know what the intent of ATHEANA is, but I
don't see that benefit with the amount of effort I have to put into it.
DR. APOSTOLAKIS: And you would recommend the committee to
write a letter stating this?
DR. BARTON: Well I think that if everybody else feels the
same way, I think we need to tell somebody, you know, maybe that they
ought to stop the process or change course or whatever.
DR. POWERS: I guess I share your concern that what we've
seen may not reveal the definite capability of this, because there seem
to be a lot of people here who are very enthusiastic about it. Based on
what was presented on the fire, I come away with -- it just didn't help
me very much.
DR. BARTON: It didn't help me either, frankly.
DR. POWERS: But putting a good face forward or seeing how
it's applied I think is something we ought to do more of and more of a
comparison to why is it so much better than the other, and I agree with
you, the fire analysis just didn't help me very much at all.
DR. APOSTOLAKIS: Mr. Siebert?
MR. SIEBER: I will probably reveal how little I know about
this whole process, but I did read the report, and I came away first of
all with a nuclear power plant perspective -- it's pretty complex; for
example and this reviews HRA, PSF, UA, HFE and HEM, all of those were
used in this discussion. For a power plant person, I have difficulty
with all of those acronyms. I had some difficulty in figuring out
ordinary things like culture and background and training, and we
struggled with that. So it could be -- the writeup could be a little
simpler as it is. The only way I could read it was to write the
definitions of all of these things down, and every time one would come
up, I would look at what I wrote down.
The second thing was the actual application. In a formal
sense, I think it's pretty good. And it would be useful to analyze some
events to try to predict the outcomes of some events from a quantitative
standpoint. That was left unreasoned. It was sort of like you arrive
at a lot of things without -- and to me, that's not quantification.
That's just a numerical opinion, and I'm not sure that that's -- the
other thing that I was struck by was when I figured the cost to apply it
would be with NUREG 2600 which was 10 to 15 people to do a level three
PRA over a period of several months.
If I add ATHEANA onto that, I basically add 5 people. I add
5 people over a period of a year or so. That's a lot of people.
Several of the people are key people, like the SRA. The training
manager; the simulator operator; I mean, our simulators are running
almost 24 hours a day at this point. So I think that the ability to
make that investment, they would have to decide who am I going to lay
off?
So there would have to be a clear description of why some of
the somebody other than the NRC would be motivated to do this, and I
can't find it in the fire scenario. There would be an awful lot of
places where it would be very, very difficult to describe, you know,
where all of this decision making or lack of decision making is. It is
understandable and logical; it's complex to read. It's the state of the
art. It would be expensive to apply. If you could show how this
benefits safety --
DR. BARTON: And improve safety?
DR. THOMPSON: And improve safety.
DR. APOSTOLAKIS: That's it?
MR. SIEBER: That's it.
DR. APOSTOLAKIS: Bob?
DR. SEALE: Well, I have to apologize first for not being
here for the presentation on fire. Mario and I were doing some other
things on license renewal. I was impressed with the fact that the
information that was presented on ATHEANA seemed to be a lot more
detailed and a lot more thoughtful than what we had heard in the past.
It's very clear that the staff has been busy trying to firm up a lot of
the areas that we had raised questions about in the past. At the same
time, I think of the 7 years. I seem to recall that it had something to
do with the cycle on some things in the Bible.
[Laughter.]
DR. SEALE: But it seems to me for all of the reasons that
you've heard from these people here and which I'm sure that you'd hear
from other people, including plant people out there plant inspectors;
that is, NRC people at the sites and so forth that you very badly need
some application to show where this process worked, and I don't know
enough about it to make a dogmatic judgment on my own as to whether or
not those applications are there, but I would advise you to look very
carefully to see if you can find someplace where you'd have a gotcha or
two, because you clearly need a gotcha.
The other thing, though, is that in terms of the things that
are in this international program, I do believe that whatever format the
human performance problem takes in the future, you can make some
recommendations as to what it takes to put our experience as we live it
today in a form which would be more readily retrievable when we do have
a human factors process that's a little more workable, and so, you know,
I just think you need to look at examples and an application. That's
where you're going to find your advocates if you're going to find any.
DR. BARTON: George, they did a fire scenario, and, you
know, if you find this thing to the Indian Point II or the Wolf Creek
draindown, what would you learn from that plant? Because I just left
the plant yesterday, and one of the agenda items we had was human
performance at the plant, and it's not improving. And I look at how
could ATHEANA really help? And when you look at the day-to-day human
performance events, this wouldn't do a thing for those kind of, you
know, day-to-day errors.
You know, you're doing control rod manipulation. This is
typical kind of stuff. You're doing control rod manipulation. You have
the guy at the controls. He's briefed; he's trained; he's licensed.
You have a peer checker. You go through the store; you go through all
of the principles. You get feedback into your three-way communications;
the whole nine yards. You're going to move this rod two notches out,
and you do everything, and the guy goes two notches in.
Now, tell me how ATHEANA -- and this is the typical stuff
that happens in a plant on a day-to-day basis. Now, tell me how I go
through the ATHEANA process, and it's going to help me do something
different other than whack this guy's head off, you know. And, see, Jay
agrees with me.
MR. PERSENSKY: They didn't get to the part of cutting his
head off.
[Laughter.]
DR. POWERS: Well, it strikes me that they will find an
approach that they could tackle exactly that question. It strikes me
that I came in here saying ah, they have a new way to do PRA, put human
reliability analysis in total in this, and I see a nice package. I
think they're not. I think they need to work on the way they tackle
really tough reliability issues.
For instance, you pretty much set up one where you could
apply all of these techniques that we talked about here to that
particular issue, and I bet you they would come up with a response. In
fact, that's the lesson I get. There is enough horsepower on it that
you will get something useful on it. And what they don't have is
something that allows me to go and do the entire human reliability
portion of a safety analysis, you know, and just turn the crank. This
is more for working on the really tough issues. It's perfect for my
surge line issue. I mean, they could really straighten Tray Tinkler
out.
[Laughter.]
DR. POWERS: Which would be a start.
MR. CUNNINGHAM: We don't want to promise too much.
MR. SIEBER: One of the things that's stated early on in the
NUREG concept is that you don't blame people, and I'm sure you want to
do that. On the other hand, when I read that, I thought secretly to
myself some people just mess up. You pull records on operators, and you
find some will make one mistake and some another, and when you move in
instead of moving out, you know, there may be a lack of attention to
detail or a lack of safety culture or a lack of attitude or what have
you that is preventing that person from doing the right thing, and I
think that you've
missed --
DR. POWERS: The documentation used to be a lot worse. I
mean, earlier documentation was really anathema to dare say that
somebody screwed up.
[Laughter.]
DR. APOSTOLAKIS: Dr. Uhrig?
DR. POWERS: I'll take another shot at it.
DR. APOSTOLAKIS: Okay; Dr. Uhrig?
DR. UHRIG: A couple of things. One, anytime I've ever been
involved with a plant with a serious problem, there has always been some
unexpected turn of events that actually changed the nature of the
problem, and I don't know how you would approach that. That's an
observation.
The second one is it strikes me that if you need data, a
modification of the LER procedures is a pretty straightforward process.
It's not simple. I don't think you go to rulemaking to get the
information that you need. I don't think so.
MR. CUNNINGHAM: It would require rulemaking, absolutely,
and a major fight.
DR. UHRIG: Yes.
MR. CUNNINGHAM: And a major fight before that rulemaking
every got very far.
DR. POWERS: I don't think that's the problem. I really
don't.
DR. APOSTOLAKIS: But if you convince people you have the
right approach --
DR. POWERS: I don't think it's a question of approach. You
know, when I first came in, you need a bunch of data to prepare this,
and I'm not sure. I think you need a bunch of problems to solve --
MR. CUNNINGHAM: Yes.
DR. POWERS: -- more than they need data to verify. I think
if I were these guys, I'd be out looking for every one of these
problems, and there's just one on the criteria for when they have to
automate versus manual action that's been sitting over like a lump, and
I think you guys could attack that problem and get something very useful
out of it.
DR. APOSTOLAKIS: Anything else?
DR. UHRIG: That issue is another one that somehow needs to
get addressed. We have literally done what we can do with training. I
think we're asymptomatically approaching this problem, well, you can
train people. Maybe automation is the next step. And I don't know
quite how this would be done.
DR. POWERS: They have a very interesting kind of plan that
would allow for people to accomplish -- you can't do it in that period
of time, you have to automate. How long do you have to rely on somebody
to recognize; they've got to do something to do it, and then, you would
surely have to -- you need those kinds of numbers, and we've got some,
you know. But there's no reason to think that it's real well-founded.
The database that they're based on is proprietary. We can't even get
it. And this looks like a methodology that I think attacks that problem
very well.
DR. APOSTOLAKIS: Dr. Bonaca?
DR. BONACA: Well, you know, thinking about what's being
done here, one of the problems I always see is about operators, people
are always writing about what the operators will do at the most distant
-- and it's very hard to bring most of this together. But, again, you
know, I want to reemphasize the fact that where it is happening in that
unique fashion was in the thinking-oriented procedure. Any experience
that has been in the industry, it was a massive experience. Only when
you put thousands of man hours when you have operators thinking together
with engineers, with people who develop event trees, very specific trees
with multiple options and so on and so forth; I think there has to be
some opportunity to benefit by grounding some of the work in ATHEANA on
comparison to what was done there, maybe just the EPGs, for example,
taking some example, getting some of the people involved in those.
I think the products will be people. You have some model of
verification. You have some way to stand on some of the hypotheses of
ATHEANA. Everything is speculative. It's probably correct, but we need
to have some benchmark.
And second, that may offer you some simplification process
and some issues that already have been dealt with in those efforts; take
a look at procedures that may -- may help you in simplifying the
process. But I can't go any further in speaking about it. But again,
the point I'm making is that that's the only place that I know operators
and analysts and development of processes came together for a long time.
But I think that there will be a great benefit, actually, in trying to
anchor ATHEANA on some benchmark, some comparison or some statement.
DR. APOSTOLAKIS: Well, I find it a bit disturbing that two
of the members with hands-on plant experience are so negative. I would
like to ask the subcommittee whether we should propose to write a
letter, whose form will have to be discussed and content.
DR. POWERS: I don't think we have to write a letter that's
critical. We need to have something that tells you to judge that data,
and I don't think we need to write a letter on the external safety
mechanisms. Cultural data, for example, on an organization.
DR. APOSTOLAKIS: The letter may say that.
DR. POWERS: If it says that, then fine.
DR. APOSTOLAKIS: Express reservations for the present state
and may urge the further application with the explicit wish that the
thing become more valuable.
DR. BARTON: I would agree with that.
DR. APOSTOLAKIS: The letter doesn't have to say stop it.
In fact, I wouldn't propose such a letter.
DR. POWERS: Maybe we should say that these people should
spend a year tackling three or four problems, visible, useful problems
that -- and show the value of this technique, because I think it's not a
technique that's going to get used. It would be wrong to hurt this,
when I think they're just getting to the point where they can actually
do something.
DR. APOSTOLAKIS: The letter, the contents of the letter are
to be discussed; I think I got a pretty good idea of how you gentlemen
feel, and certainly, I didn't hear anybody say stop this, although Mr.
Barton came awfully close.
Yes, sir?
MR. SIEBER: I wouldn't want to be interpreted as negative,
but I think things --
DR. APOSTOLAKIS: But you have been.
MR. SIEBER: No, I think things are needed.
DR. APOSTOLAKIS: Yes.
MR. SIEBER: I think simplification is needed; a good
objective is needed; what we need to accomplish.
DR. APOSTOLAKIS: Does everyone around the table agree that
a letter along those lines, which, of course will be discussed in
December will be useful?
DR. POWERS: I have reservations about the simplification,
because I know in the area -- we do have computer codes that are highly
detailed, very complex things that we use for attacking the heart of
very complex, tough problems; much more simplified techniques that we
use for doing broad, scoping analyses, and I think there's room in this
field, and I think maybe one of the flaws that's existed in the past in
this human reliability area is that everybody was trying to make the one
thing that would fit all hard problems, easy problems --
DR. APOSTOLAKIS: Right.
DR. POWERS: -- long problems, short problems, and maybe we
do need to have a tiered type of approach in which you say, okay, I've
got a kind of a scoping tool that --
DR. APOSTOLAKIS: No, I think --
DR. POWERS: I've got this one that's attacking the really
tough, really juicy problems that have defied any useful resolution in
the past.
DR. APOSTOLAKIS: I think the issue of screening, scoping
the analysis, the raised approach that was mentioned earlier, all that
part, I understand as part of this, and that was that you should have --
there also is -- but you have to convince me first that this event
deserves that treatment.
DR. POWERS: Yes.
DR. APOSTOLAKIS: And that's what's missing right now. I
would agree with Dana that you don't have to simplify everything, but
I'm inclined to say that the majority of the events would deserve it.
Now, naturally, when you develop a methodology, of course,
you attack the most difficult part, but I think a clear message here is
develop maybe a screening approach, a phased approach that would say for
these kinds of events, do this, which is fairly straightforward and
simple; for other kinds of events, you do something else until you reach
the kinds of events and severe accidents that really deserve this
full-blown approach that may take time, take experts to apply.
You know, this criticism that plant people should be able to
apply it, I don't know how far it can go, because if it's very
difficult, they are known to hire consultants. So this is the kind of
thing that they have to think about. We're not going to tell them how
to do it, but that's what I understand by your call for simplification.
You're not asking for something that says do A, B, C, and you're done.
Okay; so it seems to me that we have consensus, unless I
hear otherwise, that a letter along these lines will be appropriate to
issue, and I'm sure we'll negotiate the words and the sentences in
December.
Dana? Your silence is approval?
DR. POWERS: No, my silence is that I'm encouraging at this
point.
DR. APOSTOLAKIS: Yes; yes.
DR. POWERS: It's okay to have a methodology at this point
that only Ph.D.s in human reliability analysis can understand very well.
DR. APOSTOLAKIS: I understand the concern about the tone,
but I also want to make it very clear in the written record that these
gentlemen have reservations and not random members. I don't think Mr.
Bonaca is going to express as extreme views as you, but I'm not sure
he's far away from your thinking. So if I have the three utility
members thinking that way, I think the letter should say something to
that effect without necessarily discouraging further development or
refinement.
DR. POWERS: Yes.
DR. APOSTOLAKIS: But it's only fair; the letter will be
constructive, but it will clearly state the concerns, and perhaps we
should meet a year from now or something like that. We can say
something like that in the letter. We look forward to have interactions
with the staff.
DR. POWERS: I think I would really enjoy giving them some
time to go off and think about some problems to attack and come back and
say we think we're going to attack these two problems next time or
something like that. I think that would be really interesting, because
I think there are some problems out there that line organizations really
need some help on solving, and I'm absolutely convinced that the human
element is going to become of overwhelming importance if we're going to
have a viable nuclear energy industry in this country.
The operators are asked to do so much, and it's going to be
more and more with less and less over time, and we need to have
something that constrains us saying, yes, the operators will do this,
because right now, nothing constrains us from saying yes, the operators
have to be trained on this; they have to know this; they have to worry
about this and like that, and at some point, where that process has to
be constrained a little bit.
But I think I really come in much more enthusiastic about
this than you thought I would.
DR. APOSTOLAKIS: Okay; I think I've heard enough. I can
draft a letter. I'm sure it will be unrecognizable after --
[Laughter.]
DR. APOSTOLAKIS: But at least I have a sense of the
subcommittee.
DR. SEALE: Nobody overhead.
DR. APOSTOLAKIS: Yes.
MR. SIEBER: Can we see a copy of it before the meeting?
DR. APOSTOLAKIS: I'll do my best; I'll do my best, Jack,
before the meeting. I urge you to send emails with your concerns; yes,
and I will do my best to include your thoughts. I took notes here, but,
you know, John, if you want to send me a fax or call me.
DR. BARTON: Okay.
DR. APOSTOLAKIS: Or Jack, because I'm particularly
interested -- I mean, this is the way this committee has functioned in
the past. I mean, if cognizant members express reservations, their
views carry a lot of weight.
Is there anything else the members want to say before we
move on to safety culture?
[No response.]
DR. APOSTOLAKIS: I must say I was pleasantly surprised to
hear again the same members talk about how they wanted to see safety
culture addressed. Miracles never cease, I must say.
MR. CUNNINGHAM: Could I ask a question? I believe we're on
the agenda for the full committee in December.
DR. APOSTOLAKIS: Yes.
DR. BARTON: I think it has to be. I think after this,
you're going to have to be.
MR. CUNNINGHAM: That's the question. What would you like
for us --
DR. BARTON: To brief the other members.
DR. APOSTOLAKIS: How much time do you have?
MR. PERALTA: Probably just 45 minutes?
DR. BARTON: How much?
DR. APOSTOLAKIS: Forty-five minutes.
Would it be useful to talk about the fire scenario and in
the context of the scenario explain ATHEANA? I don't think they can do
both.
MR. CUNNINGHAM: I would agree. I don't think we can do
both.
DR. POWERS: I think they ought to just explain ATHEANA. I
don't think they should try the fire scenario.
DR. APOSTOLAKIS: I thought the scenario, the members found
extremely useful.
DR. BARTON: Well, I think yes, it is, because it shows how
they tried to apply --
DR. APOSTOLAKIS: Right.
DR. BARTON: -- the principles to an actual situation. I
think that does help. Are you sure we can't squeeze some more time off?
DR. POWERS: No.
DR. BARTON: No, we can't.
DR. APOSTOLAKIS: Let us ask if Mr. Cunningham can structure
it in such a way that he has the scenario, and on the way, you are
explaining the method?
MR. CUNNINGHAM: Mr. Cunningham will try in 45 minutes.
DR. APOSTOLAKIS: We are reminded here -- is the document
going to be available before the meeting on the fire scenario?
MR. CUNNINGHAM: I'm sorry; the --
DR. APOSTOLAKIS: We don't have anything in writing on the
--
MR. CUNNINGHAM: On the fire scenario? Will we have that
for the full committee?
MR. KOLACZKOWSKI: There is certainly a draft available.
MR. CUNNINGHAM: Okay.
MR. KOLACZKOWSKI: The NRC has not a chance to review it
yet, so it certainly is subject to revisions.
DR. THOMPSON: It's still in development.
MR. CUNNINGHAM: Okay.
DR. APOSTOLAKIS: So we're not going to have it?
MR. KOLACZKOWSKI: I don't think you're going to have it.
MR. CUNNINGHAM: No, okay.
DR. APOSTOLAKIS: Will that be a factor? We cannot comment
on something that we don't have? But we have a presentation. We have
view graphs with a comparison, so we can comment on those, right? We
can say that we didn't have a written document, but they have some nice
statements.
Mark, again, I don't want to tell you how to structure the
presentation, but the figure you have -- well, the classic ATHEANA --
MR. CUNNINGHAM: Yes.
DR. APOSTOLAKIS: -- maybe you can use that one and explain
the elements of the process and then jump into the scenario.
MR. CUNNINGHAM: Okay.
DR. APOSTOLAKIS: I don't know.
MR. CUNNINGHAM: Okay.
DR. APOSTOLAKIS: Okay? And we will try to refrain from
repeating the same questions that we have done here, right? And I see
some smiles on the faces of some of my colleagues.
[Laughter.]
DR. APOSTOLAKIS: But we will try; we will try.
I think in fairness to Mr. Sorenson, we should move quickly
on to his presentation, and I must tell you that I have to disappear at
3:30, so, Jack -- where is Jack?
MR. CUNNINGHAM: Jack is in the back.
DR. BARTON: He said 3:30.
DR. APOSTOLAKIS: But I want some discussion.
DR. BARTON: And you have to leave at when?
DR. SEALE: He has to leave at 3:30.
DR. APOSTOLAKIS: Who's leaving at 1:00?
DR. BARTON: No, I said you have to leave when?
DR. APOSTOLAKIS: 3:30. So we have about an hour and a
half. I think it should be plenty, yes?
DR. SEALE: I have to leave at about 3:30, too.
DR. APOSTOLAKIS: Okay; no problem. 3:30, 3:32.
[Pause.]
DR. APOSTOLAKIS: Okay; this is an initiative of the ACRS.
We don't know yet how far it will go; for example, our last initiative
was on defense in depth, and it went all the way to presenting a paper
at the conference PSA 1999, writing a letter to the commission and so
on. That does not mean that every single initiative we start will have
that evolution.
This is the first time that members of this committee
besides myself are being presented with this, and we also plan to have a
presentation to the full committee at the retreat; then, the decision
will be up to the committee as to what the wisest course of action will
be. We have asked members of the staff to be here, like Mr. Rosenthal,
who left; he is coming back. Jay is here, and we asked the ATHEANA
people to stay. They kindly agreed to do it. So we'll get some
reaction from experts to our initial thoughts here, and again, where
this is going to go is up to the committee, and we'll see.
Mr. Sorenson has been working very diligently on this, so I
think he deserves now some time.
Jack?
MR. SORENSON: Thank you; I am Jack Sorenson. This
discussion is based on a paper that George asked me to write earlier
this year. There is a draft on his desk for comment. But getting to
this stage took a bit longer, I think, than either one of us thought.
What I've attempted to do is put together a tutorial that
will help non-practitioners of human factors-related things to
understand what the state of the art is and what all the pieces are.
This morning, you heard -- and early this afternoon -- a great deal of
discussion on one piece of a picture that I would like to draw in
somewhat larger terms. There is no attempt here to advance the state of
the art in safety culture; just to understand it. There is no attempt
to review or critique the NRC human factors program.
What you will hear is undoubtedly a somewhat naive view, and
I would encourage those of you who are expert in one or more aspects of
the subject to offer, I hope, gentle corrections when you feel I have
misrepresented something.
DR. APOSTOLAKIS: I wonder why anyone would ask this
committee to be gentle?
[Laughter.]
DR. APOSTOLAKIS: Aren't we always?
MR. SORENSON: I was not, of course, not referring to the
committee as being ungentle.
DR. APOSTOLAKIS: Oh, I see.
[Laughter.]
MR. SORENSON: The three questions that were posed, I think
by the planning and procedures subcommittee relative to safety culture
are what is it? Why is it important? And what should the ACRS and NRC
do about it? We'll find out that the middle question, why it's
important, is probably easier to deal with than either what it is or
what people should do.
The term safety culture was actually introduced by the
International Nuclear Safety Analysis Group in their report on the
Chernobyl accident in 1986. A couple of years later, they actually
devoted a publication to safety culture, and in that publication, they
define it as shown here: safety culture is that assembly of
characteristics and attitudes in organizations and individuals which
establishes that as an overriding priority, nuclear plant safety issues
receive the attention warranted by their significance.
There are other definitions that may be useful, and we may
get to them later if it turns out that they're important, but the main
thing is that there are -- whatever definitions of safety culture you
use, there are requirements established essentially at three levels.
There are policy level requirements and management level requirements,
and those two things together create an environment in which individuals
operate, and it's the interaction between the individuals and the
environment that is generally understood to be important here.
The framework is determined by organizational policy and by
management action and the response of individuals working within that
framework. Go on to four, please.
Just a quick preliminary look at why it's important. To
understand its importance, I think you can simply look at what James
Reason refers to as organizational accidents that have occurred over the
10 years following TMI. Of course, within the nuclear industry, it was
the TMI accident that focused everybody on human factors issues. In the
10 years following TMI, there were a number of accidents where
management and organization factors, safety culture, if you will, you
know, played an important role.
The numbers in parentheses following each of these on the
list are the number of fatalities that occurred. There was an American
Airlines accident, plane taking off from Chicago where an engine
separated from the wing. It was later traced to faulty maintenance
procedures. The Bhopal accident in India, where methylisocyanate was
released resulting in 2,500 fatalities; the Challenger accident;
Chernobyl; Herald of Free Enterprise; some of you may be less familiar
with that. This was the case of a ferry operating between the
Netherlands and England that set sail from its Dutch port with the bow
doors open; capsized with somewhere around 190 fatalities.
And the last one was Piper Alpha; it was an accident on an
oil and gas drilling platform where one maintenance crew removed a pump
from service, removed a relief valve from the system, replaced it with a
blind flange which was leaking and leaking flammable condensate, and the
second maintenance crew, the second shift crew, attempted to start the
pump, and there was an explosion and resulting fire.
In the nuclear business, other than Chernobyl and TMI, we
typically end up looking at what are called near misses or significant
precursors. Two that come to mind are the Wolf Creek draindown event,
where the plant was initially in mode four, I believe; 350 pounds per
square inch; 350 degrees Fahrenheit. There were a number of activities
going on; heat removal was by way of the RHR system. There was a valve
opened, and 9,200 gallons of water were discharged from the primary
system to the refueling water storage tank in about a minute. The cause
was overlapping activities that allowed that path to be established.
There were numerous activities. The work control process
placed heavy reliance on the control room crew. There was the
simultaneous performance of incompatible activities, which were boration
of one RHR train and strobe testing of an isolation valve in the other
train. The potential for draindown was identified but was not acted
upon. Probably the most significant item here was that the test was
originally planned, the strobe testing was originally planned for a
different time and was deferred, and there was no proper review done of
the impact of that deferral.
More recent event, Indian Point II, trip and partial loss of
AC power. The plant tripped on a spurious overtemperature delta-T
signal; off-site power was lost to all the vital 480-volt buses. One of
those buses remained deenergized for an extended period and caused
eventual loss of 125-volt bus and 120-volt AC instrument bus. All
diesels started, but the one powering the lost bus tripped.
This had a number of human factors related to it. The trip
was due to noise in the overtemperature delta-T channel that was known
to be noisy, and the maintenance to fix it had never been completed.
The loss of off-site power was due to the fact that the load tap changer
was in manual rather than automatic, and that resulted in the loss of
power to the buses. The diesel trip occurred because there was an
improper set point in the overcurrent protection and an improper loading
sequence, and after that, post-trip activities were criticized by the
NRC for being more focused on normal post-trip activities and not enough
on the state of risk that the plant was in in attempting to recover from
that risk.
One of the things that is worth spending just a minute on is
the idea of culture as a concept in organizational behavior. The
International Nuclear Safety Advisory Group introduces the term safety
culture pretty much out of the blue. They make no attempt to tie it
back to the rather substantial body of literature that exists in either
anthropology, where culture is a common term, or in organizational
development, where it has become somewhat more common in the last 20
years or so.
The term is not without controversy, if you will,
particularly among the organizational development people. The term --
the idea of ascribing something called culture to an organization
started to show up in the organizational development literature in the
very early eighties. The two best-known books are probably Tom Peters'
In Search of Excellence and a book by Deal and Kennedy entitled
Corporate Cultures, and they essentially set out to determine why it was
that organizations or at least some organizations didn't behave in ways
that were clearly reflected in their structures; they were looking for
some other attribute of the organization, and they settled on the term
culture.
There are people in the literature who take exception to
that. The expectation is if you use the term culture in an
organizational sense or in the sense of a safety culture that it carries
with it some of the properties of its original use. That may or may not
be true in the case of organizational culture or safety culture, but the
fact remains that it has found a place in the literature. It is quite
widely used, particularly with respect to nuclear technology. You will
also find it in other writings in other industries, such as the process
industries and aviation.
Having said that, you find that virtually everyone then goes
on to define it in a way that suits their immediate purpose. I would
like to go back to an opening remark which I missed, and that's that I
knew I was going to have some difficulty with this assignment when I ran
across an INSAG statement that said safety culture was the human element
of defense in depth, and having spent a couple of years in defense in
depth, it just seemed unfair that --
[Laughter.]
DR. POWERS: One thing that you have to remember about the
origins of the concept is that it came up after the Chernobyl accident.
There was a strong effort among parts of some people in the IAEA to
shelter the RBNK design criticism, and you had to criticize the
operators, okay? But criticizing the operators individually was not
going to fly any better, okay? Because if you had a bad operator
individually, why did they allow it? Why did the system allow this bad
operator to be this? You had to go to this safety culture, okay?
[Laughter.]
DR. POWERS: That preserved the RBNK from being attacked,
and at the same time, it led to protecting the operators individually.
DR. BARTON: You have to admit they were a poor example of
safety culture?
DR. POWERS: What did you say?
DR. BARTON: Nothing.
MR. SORENSON: Well, that makes a good bit of sense,
obviously. Although the idea of employee attitude or management and
worker attitude having a significant impact on safety of operations, you
know, considerably predates Chernobyl. You can find references back to
the early 20th Century when industrial accidents started to become
significant in some way.
Okay; I think we can, yes, go on to -- the definition on
organizational culture, which is a little easier to deal with than
safety culture, that was offered by a critic of the Peters and Kennedy
and Deal books is the definition here. Organizational culture: the
shared values; what is important and beliefs, how things work that
interact with an organization's structure and control systems to produce
behavioral norms; the way we do things around here. This one appeared
in an article by Brill, Utah and Fortune in the mideighties, and you'll
see it repeated in very much the same form in current literature.
The last phrase, the way we do things around here, I
actually tracked back to one of the managing directors of MacKenzie and
Company. It seems to be the most concise definition of culture that --
DR. APOSTOLAKIS: That's the best one I like.
MR. SORENSON: There are competing terms: safety culture,
organizational culture, management and organizational factors, safety
climate, safety attitudes, high reliability organizations, culture of
reliability, and they all mean more or less the same or slightly
different things, depending on how they're used and what the
investigator decides to do with them.
So I think it's important to keep in mind that there are no
-- there is no generally agreed upon definition. We are dealing with
the way organizations work and the way people within those organizations
react, and at some point, you choose a definition that fits your use and
then hopefully apply it consistently thereafter.
Dr. Powers?
DR. POWERS: This one is the sixth sigma?
DR. APOSTOLAKIS: I've heard that, to.
DR. BARTON: Sick or sixth?
DR. POWERS: Sixth.
MR. SORENSON: That's one of the zero-defect --
DR. POWERS: Yes.
MR. SORENSON: -- cults, is it not?
DR. POWERS: Do everything right, yes.
MR. SORENSON: Yes; I've run across the term within the last
few weeks and I --
DR. POWERS: There was a survey in the Wall Street Journal
about a month ago.
DR. APOSTOLAKIS: Did this agency actually do a
self-assessment of its safety climate a couple of years ago?
MR. SORENSON: There was a survey by the inspector general.
I've actually been through the slides that are on the Web on that. I
don't think I've ever seen the text of the report. And they were
looking for something a little different than I would have called --
than what I would have termed safety culture. They were looking for, I
think, more of the focus of the organization on its mission and assuming
that if people were focused on the mission of the organization that that
is factory safety culture. I may be misrepresenting that but --
DR. APOSTOLAKIS: They put climate.
MR. SORENSON: They used the word culture.
DR. APOSTOLAKIS: No, I remember the word climate, because I
was impressed.
MR. SORENSON: Well, they may have used that also, but I
think the survey was titled a safety culture survey.
DR. APOSTOLAKIS: The French are using climate as well.
Climate is supposed to be really culture. Culture is more permanent,
presumably.
MR. SORENSON: One of the better-known writers in this
general field, James Reason, in his book on managing organizational
accidents lists the characteristics of a safety culture as a culture
that results in -- that encourages the reporting of problems and the
communicating of those problems to everybody throughout the
organization; a culture in which or an organizational climate in which
the members, the workers, feel justice will be done; an organization
that is flexible in the sense of being able to shift from a hierarchical
mode of operation under normal circumstances to a different mode of
operation during a crisis or an emergency and then shift back; and then,
finally, a learning organization where the information that is made
available is incorporated into the way things are done.
DR. SEALE: That clearly indicates, then, that a safety
culture is not an evolving set of values but rather a break with the
past; I mean, I can think of organizations you might characterize as a
benevolent dictatorship, and that was the way in which safety was
imposed. I guess under those circumstances, you would have to say the
old DuPont organization really didn't have a safety culture, although it
had a remarkable safety record.
MR. SORENSON: Yes; I think that's a fair characterization,
as a matter of fact.
DR. APOSTOLAKIS: And I think a lot of the old-timers in the
U.S. nuclear Navy also dismiss all of this and say Rickover never needed
it.
Now, the question is was the culture of the Navy good
because of one man? And do you want that? Or do you want something
more than that? Rickover certainly didn't think much about human
factors.
DR. POWERS: If you don't have enough people to go around --
DR. APOSTOLAKIS: I don't know, but Rickover did a good job.
DR. POWERS: There are people who would take a different
view on that.
[Laughter.]
DR. POWERS: And I think you can fairly honestly show that
there are good and bad aspects of his approach, of his tyranny.
MR. SORENSON: There were two boats lost.
DR. POWERS: The time it takes to put a boat to sea, the
mission of those boats and things like that -- you can change your
approach.
MR. SIEBER: We did a survey a number of years ago of the
idea of safety culture. About 700 people out of 1,100 responded. They
had the same list you have, except they added personal integrity to that
and caring attitude to that.
DR. BARTON: There were other characteristics.
MR. SIEBER: And that seemed to really work. It changed the
attitude in that facility; it really did, just finding out the
practices.
DR. BONACA: Although the attribute of flexibility, I think,
goes a long way in the direction. That's the key item that you
described there of when you go to technical issues, the ability of
flattening out organization and not having any more pecking order or a
fear of bringing up issues. Flexibility is very important.
MR. SORENSON: One can deduce from the literature a few
common attributes that virtually every -- all of the investigators
share: good communications; senior management commitment to safety;
good organizational learning and some kind of reward system for
safety-conscious behavior, and the lists expand from that point, if you
will.
DR. BARTON: Conservative decision making.
MR. SORENSON: I'd like to take a step back here just a
little bit and try to put the safety culture issue into the context of
the larger issue of human factors, and to do that, I think looking at
the National Research Council report done in 1988 on the NRC human
factors program is useful.
The National Research Council identified five areas that
they thought the NRC, the nuclear regulators, should address in their
human factors research. First was the human-system interface; second,
the personnel subsystem; the third, human performance; the fourth,
management and organization; and the fifth, the regulatory environment.
The first two items, human-system interface and personnel subsystem,
deal primarily with the man-machine interface, the way the machines are
designed and the way the personnel are trained.
Human performance in the context of that report is intended
to deal with what this morning was referred to as unsafe acts of one
kind or another, the actions of the system and equipment operators, and
the management and organization, what they call management and
organization factors are part of what they called a culture of --
fostering a culture of reliability. That was their phrase rather than
safety culture; and third, the regulatory environment which dealt with
the issue of how regulatory actions impacted the way the licensees did
business.
The safety culture, as I'm attempting to deal with it today,
is focused on the fourth item, management and organization. It creates
the environment that human actions are taken in, and it may contain the
ingredients to create what James Reason calls latent errors, those
things which change the outcome of an unsafe act, but the issue of
safety culture deals with the management and organization factors and
the climate it creates, the conditions it creates for the human to
operate in.
One of the difficulties I had in going through the
literature was trying to understand what all the pieces were, and so,
one of the things that I ended up doing that helped me, and I think
could be generally helpful in putting some of the pieces together is to
look at all of the things that go into the process of establishing some
interesting relationship between something called safety culture and
operational safety or ultimately some measure of risk, and this figure
shows the first half-dozen steps in that process.
But the idea here is if safety culture is interesting for me
from an operational safety standpoint, you need to be able to establish
something about those relationships. The process typically starts off
with defining some kind of an organizational paradigm. Mintzberg's
machine bureaucracy is very often used for nuclear power plants, and
then, as soon as it's used, it's criticized for having several
shortcomings. The investigators need to have some idea of how the
organization works, and they generally should start with some definition
of safety culture, what it is.
Having done that, then, they need to define some attributes
of safety culture, and it might be the ones that I listed a few minutes
ago: good organizational learning, good communications and so forth,
but there are somewhere between a half a dozen and 20 of those
attributes that can be identified, and having done that, then to
evaluate organizations, you need to look -- you need to have a way to
measure those things that you've just identified, and you might put
together personnel surveys or, you know, technical audits or whatever,
but you need some kind of evaluation technique that involves looking at
how the organization, how an organization actually works.
Having designed the evaluation techniques, you need to
collect data, and then, you need to have, once you have data, you need
to have something to -- that tells you how to judge that data, and I've
indicated that by choosing external safety metrics; if you collect
cultural data, for example, on an organization, how do you decide that
that organization is safe or not safe in judging the cultural data?
In their simplest form, those external metrics might be a
SALP score. They might be the performance indicators that we're using
now. They might be earlier performance indicators. But the
investigator makes some choice of what he's going to compare his
cultural parameters to.
And typically, that correlation is done with some sort of
regression analysis, and as a result of doing the, you find out that
some number of the safety culture elements you started with, you know,
correlate with your safety parameters, and some don't. And the output
from that first stage, then, is which of these safety culture elements
turn out to be significant.
The remainder of the process, then, if you want to carry it,
you know, all the way to its logical conclusion is you would like to be
able to use these significant safety culture elements to modify in some
way your measure of risk, and the next figure -- if you can move that
one over a bit; pick up the balance of that. The bottom path there
identifies, you know, relating the elements that you've decided are
significant to the PRA parameters or models; box 11 finally modifying
the PRA parameters and ultimately calculating a new risk metric.
DR. APOSTOLAKIS: So I guess ATHEANA, then, because you
don't necessarily have to go to that, ATHEANA would be somewhere there
in between 9 and 10, perhaps?
MR. SORENSON: I would put -- well, it doesn't work on
performance indicators, as I understand it. I would say ATHEANA covers
8 and 11; is that a fair statement?
DR. APOSTOLAKIS: It definitely does, but perhaps to take
advantage of the qualitative aspects, you need an extra box so you don't
just make it PRA. So before eight, you might have the qualitative
aspects of ATHEANA, and then, at the start of eight, of course, you have
to do the quantification.
MR. SORENSON: I would be delighted to get critiques on
this, too.
DR. APOSTOLAKIS: Don't worry; don't encourage people.
[Laughter.]
DR. APOSTOLAKIS: Susan, you wanted to say something? You
have to come to the microphone, please; identify Your Honor.
MS. COOPER: Susan Cooper with SAIC.
I think with respect to interaction with ATHEANA, there are
certainly two different ways. Already, we're trying to incorporate some
symptoms, if you will, of culture and some of the preparation for doing
ATHEANA. We'd like the utility people to try to examine what are their
pre-operational problems as part of identifying what their informal
rules or maybe some things that are, if you will, symptoms of a culture
that, when they play it out through a scenario development and
deviations, it would be organizationally-related, but we don't have what
we see from some of the events, some of the other things that the
organization can do that might set up a scenario, so we recognize that
there may be some pieces missing, and we certainly need some kind of
input to know not only what -- you know, what from the organization is
going to cause things but then also, then, what is the impact on the
plant? There are a couple of different pieces.
DR. APOSTOLAKIS: Now, if just for a couple of historical
purposes, we go to the previous one, no, yes; box four, collect and
analyze data, that was essentially the reason why one of the earlier
projects on organizational factors funded by this agency was killed.
The proposed way of collecting data was deemed to be extremely
elaborate. They implemented it at Diablo Canyon, and the utility
complained.
So, there is this additional practical issue here that you
have to do these things without really --
DR. POWERS: I don't know why.
DR. APOSTOLAKIS: Dana's commentary here, I mean, certain
things, by their very nature, require a detailed investigation. I mean,
I don't know where this idea has come from that everything has to be
very simple and done in half an hour, but I think it's important to bear
in mind that the utility complained, and the management of the agency
decided no more of this. I'm willing to be corrected if anybody knows
any different, but that was my impression.
MR. SORENSON: Well, and we'll touch on that
one --
DR. APOSTOLAKIS: Okay; sorry.
MR. SORENSON: -- a little in a couple of slides, as a
matter of fact, but you're right: one of the results early on was that
people did try to look for non-intrusive ways to collect data. One
possibility is to look at the way the organization is structured, which
you can deduce from, you know, organizational documents, if you will.
DR. APOSTOLAKIS: Yes, but the attitudes, you would never
get that.
MR. SORENSON: You don't pick them up and --
DR. APOSTOLAKIS: These attitudes, you don't pick that up.
MR. SORENSON: And interestingly enough, the people that
started down that path after a few years started to pull in something
that they called culture, the way an organization worked.
Yes; I will, time-permitting, go through at least one
example that sort of traces through those boxes, if you will. I would
like to comment on the upward path on slide 16. The -- what you would
really like to do is to be able to identify some number of performance
indicators that were indicative of the safety culture elements and that
you could translate, in turn, into modifications of the PRA parameters,
and the idea there is if you can identify those performance indicators,
then, you don't have to go back and do the intrusive measurements once
you've validated the method.
And so, in the best of all possible worlds, you know, one
would, you know, have processes that follow that upward path. Now, I
would hasten to add in summarizing on this figure that there is a lot
that goes on inside every one of those boxes, and, in fact, when I was
discussing this with Joe Murphy -- I guess he's not here today -- and at
one point, we pointed at one box in particular, and I asked him a
question about it, and he said, well, of course, in that box, miracles
occur, and that's still --
DR. APOSTOLAKIS: Did he also tell you that there's a NUREG
from 1968 whose number he remembered that addresses it?
[Laughter.]
DR. APOSTOLAKIS: I mean, Joe usually does that.
[Laughter.]
DR. APOSTOLAKIS: PNL published a report in 1968 in March --
[Laughter.]
DR. APOSTOLAKIS: -- that is relevant.
MR. SORENSON: So the -- anyway, the summary here is that
this path is neither short nor simple.
DR. APOSTOLAKIS: Yes.
MR. SORENSON: There are a lot of pieces that go into
establishing a relationship between safety culture or other management
and organizational factors and some risk metric.
Let me see what we might need to do here. How much time do
you want to leave for discussion, George?
DR. APOSTOLAKIS: Well, you are doing fine.
MR. SORENSON: Okay.
DR. APOSTOLAKIS: I think people can interrupt as they see
fit.
MR. SORENSON: Okay.
DR. APOSTOLAKIS: So, you're doing fine.
MR. SORENSON: What I'd like to do now is go through some of
the boxes and some examples of some work that has been done referring
back to figures 15 and 16. As the figure indicates, the process starts
out somehow with a model of the organization you're interested in, and
my conclusion as a layperson was that you can look at essentially the
way an organization is structured; the way it behaves or its processes
or some combination of those things.
If you look at slide 18, this was an attempt to look at
structure only. The work actually started at, I believe, Pacific
Northwest Laboratories and was continued by the same investigators,
although at different places, over the next several years, and here,
they attempted to look strictly at what they could deduce from the way
the organization described itself, if you will. It does not involve
culture. If you look at the literature referenced by these folks versus
the literature referenced by organizational culture people, it's a
different body of literature. There's very little cross-referencing.
This was designed to be non-intrusive. It has an obvious
difficulty right up front, and that is that there are a lot of factors
to try to correlate. They made an attempt to correlate with things like
unplanned scrams, safety system unavailabilities, safety system
failures, licensee event reports and so forth. There was other work
sponsored by the NRC that began at, I believe, at Brookhaven; Sonia
Haber and Jacobs and others, not all at Brookhaven, I would hasten to
add, and this was a slightly different perspective on the same thing.
They came up with 20 factors that included something they called
organizational culture and safety culture, and this was the -- where the
-- one where the data gathering, if you will, did become very intrusive.
They made up surveys and went out and talked to a bunch of plant people
and shadowed managers and so on and so forth, and they probably got
pretty good data, but it was not an easy process.
Then, there is another process developed by -- I was going
to say that eminent social psychologist.
DR. APOSTOLAKIS: I would like to add that Mr. et al. is
here.
MR. SORENSON: Yes, good.
DR. APOSTOLAKIS: His first name is et; last name is al.
[Laughter.]
DR. APOSTOLAKIS: We call him Al.
MR. SORENSON: Anyway, one of the contributions here was to
reduce the 20 factors to half a dozen, which makes the process more
tractable, if you will, but it's a little different also in the sense
that it focuses on the work processes of the organization and how those
are implemented, and, in fact, the next figure, I believe, is an example
of their model of a corrective maintenance work process, and the
analysis includes looking at the steps in the process and identifying
the -- what they call barriers or defenses that ensure that an activity
is done correctly, and you can map these activities back onto the
earlier list of six attributes, if you will, to determine the
relationship between the organization and the work processes.
DR. APOSTOLAKIS: One important observation, though: these
six are not equally important to every one of these. This is a key
observation. For example, goal prioritization really is important to
the first box, prioritization of the work process, whereas technical
knowledge, for example, means different things for execution and
different things for prioritization. So that was a key observation that
Rick made on the factors that Haber and others proposed to deal with the
work process. Then, it meant different things than he proposed.
And most of the latent errors of some significance were the
result of wrongful prioritization. That is, we will fix it at some
time, when it breaks; unfortunately, it breaks before you could --
MR. SORENSON: Okay; moving on to the next box in the
activity diagram, coming up with some way to measure safety culture or
whatever organizational factor you are concerned with, there is, you
know, the obvious candidates: document reviews, interviews,
questionnaires, audits, performance indicators. But I think the thing
that struck me here is that regardless of what list of safety culture
attributes you start with, in this process, you're going to end up with
some questions that you hope represent those attributes in some way, so
when you get done, you don't have just, you know, a direct measurement
of organizational learning; you have answers to a set of questions that
you hope are related in some way to organizational learning.
DR. POWERS: The difficulty in drafting the questionnaire
that gives you the information that you're actually after must be
overwhelming. I mean, the problems that they have on these political
polls, they can get any answer they want depending on how they construct
the question. I assume that the same problems affect the
questionnaires.
MR. SORENSON: I would assume so, but this is also to assume
what psychologists -- the organization --
DR. APOSTOLAKIS: Never rely on one measuring instrument.
MR. SORENSON: Would anybody like to comment on the
difficulty that goes on within that box?
DR. APOSTOLAKIS: It's hard.
MR. SORENSON: It's hard.
DR. POWERS: That's a separate field of expertise,
formulating questionnaires, is it not? I'm really concerned that you
asked too much to be able to formulate a questionnaire that allows
somebody to map an organization accurately when you have this difficulty
that I can get any answer that I want depending on how I construct the
questions.
MR. SORENSON: Of course, part of the way round that is --
well, there are ways of designing questionnaires so that the same
question gets asked six different ways, and you can check for
consistency and poor wording.
DR. POWERS: What do you do when they're inconsistent? Do
you throw it out?
MR. SORENSON: That's what you pay psychologists for.
DR. POWERS: I mean, I don't see that you're out of the game
here. I mean, I had enough to do with employee opinion poll taking and
what not that it's been known that there is a culture or a discipline
doing these things, and there are well-known principles, like the second
year of the employee opinion poll, the results are always worse than the
first year; the people filling out the questionnaires have gotten better
at filling out questionnaires, so they can be more vicious in their
evaluations. I mean, it just strikes me as a flawed process.
MR. SORENSON: Well, I think part of the answer to that is
you try to measure enough things that if your measure is flawed on one
or two or three of them, you can still get the -- an indication of the
attribute that you're really trying to measure.
DR. SEALE: It's interesting, because so many organizations
now have been convinced that their organization has to be a
participatory autocracy, and so, they ask these questions in the
questionnaires, and as you say, they deteriorate almost invariably, but
they also systematically ignore the results, so that --
[Laughter.]
DR. SEALE: But, you know, in the name of, as I say,
participatory autocracy, they do it.
DR. POWERS: I am intimately familiar with one organization
who is absolutely convinced that the fact that they conducted a
questionnaire on a particular aspect of behavior excuses them from ever
again having to attend to that.
[Laughter.]
DR. APOSTOLAKIS: Why didn't you include the behaviorally
anchored rating scales?
MR. SORENSON: I didn't intentionally exclude it. I didn't
see it as different from -- in a process sense from what's here. I may
have misread that.
DR. APOSTOLAKIS: Anyway, okay, that's another of the
instruments that's available.
But let's go.
MR. SORENSON: Okay; selecting external safety metrics: I
mentioned that briefly earlier, you know, one can rely on performance
evaluations, performance indicators, do some sort of expert elicitation
to evaluate the organization. In some industries, which we'll touch on
in particular, process in aviation, actually, I have accident rates that
you can use as a metric, where there is good statistical data on
accident rates. But again, the point I'm trying to make here is that
the investigator chooses that as part of the evaluation process, and
sometimes, that is lost sight of.
In the chemical industry, process industries in particular,
they tend to use the audit techniques. They don't have the same
reluctance to gather field data that seems to exist in the nuclear power
business. They tend to use the terminology safety attitudes and safety
climate versus safety culture, and the studies that I've looked at used
either self-reported accident rates or what they call loss of
containment accident rates, you know, covering relatively large numbers
of facilities. One study covered, I think, 10 facilities managed by the
same company, for example; 10 different locations.
And these studies in the process industries have resulted in
very strong statistical correlations between the attributes of safety
culture that we've been talking about here and accident rates, and you
can show that the low accident rate plants, you know, show strong safety
culture attributes. The typical correlation they might start out with,
you know, 19 or 20 attributes as the Brookhaven people did and find out
that 14 or 15 of those correlate and five don't for some reason.
DR. SEALE: Jack, how much of that, though, is due to the
fact that the elements of positive numbers on the accident rate are the
inverse or one minus the numbers on the safety culture? I mean, they're
almost -- the way you characterize your safety culture almost certainly
is painted by the idea that one of the worst things that can happen to
you is an accident.
MR. SORENSON: Well, certainly, you've got to look at how
the measurement is done. I don't have a quick answer.
DR. SEALE: No, I mean, what if you had just for instance or
just for the fun of it, let's say we had two plants, and both of them
didn't have any accidents; one of them had a good safety culture and one
of them didn't. I don't know if your questionnaire would actually
detect or make that distinction.
MR. SORENSON: In that case, I think you're absolutely
right, but precisely the point I'm trying to make here is that in this
case, we are not looking at plants with zero accident rates. We're
looking at plants that have very low accident rates and very high ones.
DR. SEALE: Yes.
MR. SORENSON: So we've got statistics here that we don't
have in the nuclear power business. The ratio of the best performing to
the worst performing in terms of accident rates is typically about 40,
the factor is. And, in fact, I'll come back to that later. The reason
that one of these folks makes the point is in aviation --
DR. APOSTOLAKIS: PSA is one minus the -- you know, that's
my problem.
MR. SORENSON: The aviation business, which presumably uses
roughly the same equipment and roughly the same training methods
worldwide for commercial passenger airlines, there's a difference of
about a factor of 40 between the best and worst performing airlines.
DR. SEALE: Yes.
MR. SORENSON: So the point here is precisely that in those
areas where you've got data, you can correlate these safety culture
elements, if you will.
Which brings us to, you know, the areas of weakness or
discomfort, most of which have been touched on here earlier. One of
them is that at this point, nobody pretends to understand the mechanism
by which the thing we call safety culture affects operational safety.
Second area was what you just touched on, Bob. There is a lack of valid
field data in the nuclear power business in particular. First, the
actual accident rates are low, but there's even a lack of data on the
safety culture side in general.
And the third area is there are no good performance
indicators that have been identified at this point; clearly an area that
needs additional attention, not only in the nuclear power business.
DR. BARTON: I think you're looking at too high a level for
the field data to be looking at accidents. I think you don't have to
look at accidents. Go look at lower levels of performance in the
organization; go look at industrial safety events. Go look at human
performance or look for operator errors. Go look at maintenance people
not following procedures.
If you go look at a whole bunch of those things and relate
that, you'll find out that the culture is different at that plant than
it is at the other plant that hasn't had a major accident either but
doesn't have the same numbers of those types of --
DR. SEALE: You could probably use LERs just as easy of
that.
DR. APOSTOLAKIS: Or any number of attributes --
DR. BONACA: The trouble with LERs is there are not enough
LERs written. These plants write three or four LERs a year. I don't
know if there's enough data there.
DR. BARTON: Or whatever the correct level of --
DR. SEALE: Yes.
DR. BONACA: There are corrective action systems at the
plants --
DR. SEALE: Yes, yes.
DR. BONACA: Because there are 20,000 inputs per plant.
DR. SEALE: Yes.
DR. BONACA: Probably, that's the biggest window that you
have.
DR. APOSTOLAKIS: So you are saying that it would be perhaps
worthwhile to see if some performance indicators can be formulated using
this kind of evidence?
DR. BARTON: I think so.
DR. APOSTOLAKIS: Instead of going to models? That's a good
idea.
DR. BARTON: Think about it.
DR. APOSTOLAKIS: It would be extremely tedious to go
through those records.
DR. BARTON: Oh, yes.
DR. APOSTOLAKIS: But it would probably be worthwhile.
MR. SIEBER: A lot of plants.
DR. POWERS: You can find people within an organization
oftentimes who know those records surprisingly well. If you have a lot
more, then it's a lot easier.
DR. BONACA: I mean, an example of performance indicators at
IAEA and all places, one could ask whether or not they should be nine or
whatever. But have they had those elements that were --
DR. SEALE: They weren't accidents.
DR. BONACA: No, incidents.
DR. APOSTOLAKIS: But that is a necessary assumption that
this really is a good indication of what will happen if there is a need
for an ATHEANA kind of system, but it may be very good when it comes to
a major -- when they pay attention. In fact, we had a guy call
maintenance people; more than 50 percent, to my surprise, thought that
the procedure was useless; they never followed them. They thought they
were for idiots.
Now, those guys probably are very good, but if you are
blind, you say oh, they don't use the procedures; my God, bad, bad boy.
Yes; they're probably doing a better job than somebody else who goes
with --
DR. BONACA: Even there, that's another issue.
DR. APOSTOLAKIS: So I think there is this presumption,
although I like the idea, because at least you get something concrete,
but maybe that's something else to think about: how much can you
extrapolate from these fairly minor incidents, because there is this --
Jack didn't mention, but people also distinguish between the formal
culture and the informal culture, the way things really get done. And
do they take shortcuts? They do all sorts of things. And these are
good people usually. I mean, they're not -- but I think that's a good
idea. It's a good idea. It's just that, I mean, they have -- you know,
whenever anybody proposes anything here, you have to say something
negative about it. So, there it goes.
Alan, you have to come to the microphone.
MR. KOLACZKOWSKI: Alan Kolaczkowski, SAIC.
George, that's the very reason why, in the ATHEANA part, I
think we're looking at both the EOPs and the formal rules, but then, you
saw we also look at tendencies and informal rules.
DR. APOSTOLAKIS: Right.
MR. KOLACZKOWSKI: That's where we're trying to capture some
of those -- part of the culture, if you will: how do they really do it?
What are the ways they really react when this parameter does this? What
are their tendencies? I think we're trying to capture some of that. We
use the terminology informal versus formal rules, but I think we're
talking about the same kind of thing.
DR. APOSTOLAKIS: Yes.
MR. SORENSON: By the way, though, not all investigators
agree that let me call them near misses or incidents extrapolate
properly to accidents.
DR. APOSTOLAKIS: Yes, you have to make some assumptions.
MR. SORENSON: And also, the people who question that also
question whether the human performance information or models in the
nuclear business translate to those in other hazardous industries.
That's not a given.
DR. APOSTOLAKIS: Go ahead.
DR. SEALE: But the point may be, though, that the extent to
which the organization has the capability of absorbing near misses in
such a way that they do not propagate to major accidents may be the
thing that's the measure of safety culture.
MR. SORENSON: Well, Reason would agree with that very
precisely, because his definition of safety culture, you know, is, in
effect, that culture which leads to a small incidence of latent errors
that go undiscovered. And it's the latent errors that translate, you
know, a single unsafe act into a disaster.
DR. SEALE: And then, but the ability to correct for the
error in other parts of the organization so that it doesn't grow --
MR. SORENSON: Right.
DR. APOSTOLAKIS: But I think another measure of goodness
which is really objective is to see whether they actually have work
processes to do some of these things. Rick is working with -- Rick Weil
is trying to develop organizational learning work. So what you find is
that yes, everybody says, boy, organizational readiness, sure, yes, we
do that. But how do you do it? And that's where he gets stuck. We do
it. Somehow, we do it. There is no work process; they have no formal
way of taking a piece of information, screening it, because that's the
problem there: they get too many of those.
DR. BARTON: How many do they get a week or about a year?
DR. APOSTOLAKIS: About 6,000 items a year; I mean, here,
you're not going to be producing power just to study 6,000 items.
[Laughter.]
DR. BARTON: I hope not.
DR. APOSTOLAKIS: So there is no formal mechanism for
deciding what is important, which departments should look at it, and I
think that's an objective measure.
DR. BARTON: Yes, it is, because you can prioritize those
6,000.
DR. APOSTOLAKIS: But they don't.
DR. BARTON: You can put them in buckets. Well, I know
plants that do.
DR. APOSTOLAKIS: I'm sure; and those have a better culture.
DR. BARTON: I don't necessarily agree with that.
[Laughter.]
DR. APOSTOLAKIS: All right; no, but it is an objective
measure of the existence of the processes themselves. It is a measure
of some attempt to do something.
DR. BONACA: But it is also a measure of the way the work is
getting accomplished or not accomplished that gives you some reflection
on potential initiators. For example, a process that is overwhelmed
that is unable to accomplish work on a daily basis, something is going
to happen out there, because we're starting an item; you are closing it.
You're delaying items, and something is going to start in a new activity
before you close the other one at some point.
And so, if you look at that, you have a clear indication,
and we're trying to begin to correlate that. So you have some
indication of really what kind of a story. Now, the question is are
they going to affect the unavailability of a system? See, we don't know
that.
DR. APOSTOLAKIS: It may, but -- but there is something to
the argument that -- not just nuclear. But it seems to be consensus of
organizational learning is a key characteristic of good organizations.
Now, if I see that, I really don't need to see real data to prove that.
I mean, those guys are not stupid. They know what they're talking
about. And, in fact, I remember there was a figure from a paper in the
chemical, whatever; it was a British journal, comparisons of good
organizations, excellent organizations. The key figure that
distinguished excellent from everybody else was this feedback loop,
organizational learning, from your own experience and that of others,
and it's universal.
Anyway, let's have Jack continue. He's almost done, I
understand.
MR. SORENSON: Yes; there are a couple more slides here, and
I did want to touch on what we've just been discussing, you know, the
evidence that a safety culture is important to operational safety.
There is an overwhelming consensus among the investigators; if there is
a subculture that thinks an attitude doesn't matter, I didn't find it in
the literature in any event.
The accident rate data is pretty convincing. I confess
obviously to not being an expert, but the writing, again, supports that.
People outside of the field seem to think they have good statistical
information there. And the little bit of nuclear power plant field data
that there is, some of what the Brookhaven people did, Hauber and her
colleagues and the little bit that was done in the Pacific Northwest
Laboratory work confirmed a correlation between safety culture elements
and operational safety as they defined it. There are not enough data,
but what's there was positive.
I'm going to, on the last slide, relate my impressions again
as a non-practitioner as to what is missing from the literature. Some
of this, I've deduced from what other people have written and some just
from my own feelings on the papers that I review. There is a lack of
field data relative to nuclear power plant operations. There might be
easy ways to get it, but right now, it's not there.
One needs to understand the mechanism by which safety
culture or other management and organizational factors affect safety.
We need performance indicators for safety culture or related things. We
need to understand the role of the regulator in promoting safety
culture, and we need to know something about the knowledge, skills and
abilities of the front line inspectors in a regulatory environment where
safety culture is important. One of the things that struck me in doing
the research on this work is that we are -- we, the NRC -- are right in
the middle of attempting to change the way we do regulation. We are
embarking on and evaluating a new reactor oversight process. We are
trying to convert our regulatory basis to something we're calling
risk-informed and maybe performance-based, and other regulators
elsewhere in the world, particularly in the UK, are observing that. If
one is going to make this kind of a change, then you probably cannot do
it within the kind of prescriptive regulatory framework that the U.S. is
using at the moment.
That being the case, something called safety culture and how
one fosters it becomes very important relative to the new regulatory
process that we are expecting to implement. There is certainly a
reluctance on the part of the NRC to, you know, venture into anything
that would smack of regulating management and an even stronger
reluctance on the part of the industry to, you know, allow any small
motion in that direction, but it seems to me that in the context of this
new regulatory regime, that management is terribly important, and at a
minimum, the agency needs to understand in what ways is it important,
and how does the agency best foster this ownership of safety amongst its
licensees, and I don't think we know that right now.
That's all I have.
DR. APOSTOLAKIS: Yes; the big question is really what is it
that a regulator can do without actually managing the facility. That's
really the fear.
Dennis Bley, please?
MR. BLEY: My name is Dennis Bley. I'm with Buttonwood
Consulting. I have to leave in just a minute --
DR. APOSTOLAKIS: Sure.
MR. BLEY: -- so I thought I would say a couple of words
quickly. The last 5 years, I've been on the National Academy committee
overseeing the Army's destruction of chemical weapons, and the program
manager for chemical weapons destruction has sponsored a lot of digging
into this area, and I think maybe they would be willing to share what
they've found.
We've had people on our committee from DuPont, and, you
know, the strong view from DuPont, coming back to what you were talking
about earlier, is that if you get the little things under control, the
industrial accident rates, those things, you won't have a bad accident.
A lot of people don't believe that. They do very strongly. Jim
Reason's book you were talking about, I think the last chapter, tenth
chapter, he goes into that in some detail.
I kind of think from NRC's point of view, it gets difficult,
because the expertise the Army has brought together to help them look at
this in many places has all argued strongly that strong regulation and
compliance don't get you where you want to be with respect to safety; it
has to be the individual organization taking ownership, and all the way
through, certain things are unacceptable, certain kinds of behavior are
unacceptable by anybody, and that has to get buried into the whole
organization.
Just an aside on ATHEANA, it would be -- where you pointed
out where they would fit together, I think that's about right, and we've
actually got, if you look at some of our examples, a little of that
coming in but nothing like a real solid process for trying to find all
of it. But I think you can -- there has been so much work in this area
by so many different people, including studies in industrial facilities,
that it probably doesn't make sense to do it all over again.
But I'll just leave it with that. That's the one source
I've seen where people have -- they've really tried to draw a broad
range of expertise together to help them with the problem, which they
haven't solved.
DR. APOSTOLAKIS: I believe the fundamental problem that we
have right now is that people understand different things when the issue
of culture is raised and so on. There was a very interesting exchange
between the commissioners and the Senators. Senator -- I don't
remember; Inhofe?
DR. SEALE: Inouye?
DR. APOSTOLAKIS: No, no, no.
DR. SEALE: Inhofe, yes.
DR. APOSTOLAKIS: He was told by Former Chairman Jackson
something about -- it was somebody else; not the chairman about culture
and organizational factors and boy, he said I've never heard -- he said
I'm chairing another subcommittee of the Senate where we deal with
Boeing Corporation and all of those big -- and I've never heard the FAA
trying to manage the culture at Boeing and this and that, and how dare
you at the NRC think about that?
And then, of course, we have our own commission stopping all
work, you know, overnight a year or so ago, and I think it's this
misunderstanding; you know, I really don't think it's the role of the
regulator to go and tell the plant manager or vice president how to run
his plant. On the other hand, there are a few things that perhaps a
regulator should care about. I don't know what they are, but for
example, the existence of a minimum set of good work processes, in my
opinion, is our business, and especially if we want to foster this new
climate that I believe both Dennis and Jack referred to. In a
risk-informed environment, some of the responsibility goes to the
licensee. Now, we are deregulating electricity markets and so on, so
that's going to be even more important.
But I guess we never really had the opportunity to identify
the areas where it is legitimate for a regulatory agency to say
something and the areas where really it is none of our business, and
it's the business of the plant. And because of the fear that we are
going to take over and start running the facility, we have chosen to do
nothing as an agency.
DR. SEALE: Well, that goes to the question of where is it
we ought to butt out? Where should we butt out? What are the things
that we do that are counterproductive?
DR. APOSTOLAKIS: Absolutely right; absolutely right.
DR. BONACA: But again, I think if you want to talk about
culture, management up there, it's very, very hard, and again, we're
struggling with looking at an indication of an organization that works
or doesn't work. At the industrial level, there are indications all
over the place. But those indicators have to do with does the work
process work, for example? Is the backlog that people perceive they
have overwhelming them? What kind of -- absolutely. And again, there
is work that is being done inside these utilities to look at those
indicators there, and they don't even measure management per se; simply
something is wrong with the organization. When you have something wrong
with the organization, you go to the management, and you change it,
because you expect that you will be able to manage that.
But I'm saying that it's probably feasible to come down to
some of these indicators, and I think that the utilities are trying to
do that.
MR. SIEBER: I would sort of like to add: I've been to some
regional meetings for clients of mine where the plants have been having
problems, where the regional administrator or his staff has asked
questions about performance indicators on productivity, and for example,
a lot of these processes are just a bunch of in-boxes, you know, like
your work process. Which one is the in-box that has big holes in it?
Why isn't work getting done? I've seen the NRC ask those questions. I
think they're legitimate questions, and on an individual basis, I think
that they're appropriate questions, but I have not seen an initiative to
ask them across the board.
DR. BARTON: They all do relate to cultural issues.
MR. SIEBER: That's right.
DR. POWERS: Yes.
MR. SIEBER: Each one of them by itself is an indicator, and
I think industrial safety is a prime indicator. You know, if you --
DR. BARTON: If it wasn't, they wouldn't spend so much time
looking at it.
MR. SIEBER: Yes, and we actually hired DuPont, who is very
good, to help us with ours, and our record, our accident rates, went
down by over 90 percent. I mean, it actually worked, and that's part of
the culture. If you can't make yourself safe, how can you make a power
plant safe?
DR. BARTON: There are things you can look at without really
getting into the management, so to speak, of the company. I think you
have to draw that line, because the industry is going to get nervous as
heck. They're just going to say -- they'll start looking at the safety
culture and management's confidence and all that stuff. I think there
is a set of things that you can look at objectively and determine what
is the culture of this organization. You just have to figure out how to
package it.
DR. APOSTOLAKIS: That's the problem.
DR. BARTON: How to package it.
DR. APOSTOLAKIS: That's the problem.
DR. BARTON: Expect that if you're looking at a bunch of
indicators right now that I would tell you would fit into a box called
culture. Look at it right now.
DR. BONACA: Well, I mean, again, there have been efforts;
I've been participating in one, and I believe that if you look at other
people who do it, they're finding out the same points. Now, again,
you're going down to opinions for objective readings of certain boxes of
work being accomplished or not accomplished.
DR. BARTON: And that's the problem. It's what you can do
when you take this data, and you get it back to the region, and that's
where people really get nervous now.
DR. BONACA: But I was talking about trying to correlate,
for example, working efficiencies of backlogs, actual outcomes that you
can measure somewhat for using PRA. That's -- I mean, that's probably
something that you can do.
MR. SIEBER: One of the problems is that the boxes from
plant to plant are not standardized. The thresholds that differ from
plant to plant. So interplant comparisons are not very accurate. On
the other hand, you know, something is better than nothing. And that's
what plant managements use to determine the state of culture and how
safe they are and how safe they aren't and how well their processes
work. That's how you run the plant.
DR. POWERS: One of the things that I find most troublesome
right now is taking the DuPont experience, and this attitude I hear all
the time, the Mayer approach toward safety; you take care of all the
little things, and the big things will take care of themselves versus we
want to focus on the most important things in risk assessment. We seem
to be dichotomizing opposite views. I'm wondering if we really want the
outcome we're going to get going to risk-informed.
DR. SEALE: I'm not so sure.
DR. POWERS: It seems like it's worth thinking about,
because these things have been very successful in another industry.
DR. SEALE: The thing, though, is that the things that are
getting ruled out, if you will, on the basis of not contributing to risk
are not the little things that show up in the plant performance things.
They're truly the -- they're the not even on the radar screen things.
At least that's my impression. It's a good point, but I don't think
you're talking about the same population when you say risk versus low
risk on the one hand and little things versus big things on the other
hand.
DR. APOSTOLAKIS: Anyone from the staff or from the audience
want to say anything?
[No response.]
DR. APOSTOLAKIS: Okay; any other comments?
[No response.]
DR. APOSTOLAKIS: Thank you very much. We will adjourn.
So, this meeting of the subcommittee is adjourned.
[Whereupon, at 3:10 p.m., the meeting was concluded.]