ACRS/ACNW Joint Subcommittee - January 13, 2000
UNITED STATES OF AMERICA
NUCLEAR REGULATORY COMMISSION
ADVISORY COMMITTEE ON REACTOR SAFEGUARDS
MEETING: ACRS/ACNW JOINT SUBCOMMITTEE
11545 Rockville Pike, Room T-2B3
Thursday, January 13, 2000
The subcommittee met pursuant to notice, at 8:30 a.m.
THOMAS KRESS, ACRS, Co-chairman
JOHN GARRICK, ACNW, Co-chairman
GEORGE APOSTOLAKIS, ACRS, Member
RAYMOND WYMER, ACNW, Member
. P R O C E E D I N G S
MR. KRESS: Could we please come to order?
This is a meeting of the Joint Subcommittee of the Advisory
Committee on Reactor Safeguards and the Advisory Committee on Nuclear
I am Thomas Kress. I'm co-chairing this joint subcommittee,
and on my right is Dr. John Garrick, who is the other co-chair of the
I guess I'll be mostly in charge of this particular meeting.
Other joint subcommittee members in attendance are Dr.
George Apostolakis of the ACRS, Dr. Ray Wymer of the ACNW, and also
present is Dr. Milt Levenson, who is a consultant to the ACNW.
The purpose of this meeting is for the joint subcommittee to
discuss the defense-in-depth philosophy in the regulatory process,
including its role in the licensing of a high-level waste repository,
its role in revising the regulatory structure for nuclear reactors, and
how the two applications should be related to each other.
The discussion will also include the role of
defense-in-depth in the regulation of nuclear materials applications and
other related matters.
The subcommittee will gather information, analyze relevant
issues and facts, and formulate proposed positions and actions, as
appropriate, for deliberation by the full committees.
Michael Markley is the designated Federal official for the
initial portion of 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 December 21, 1999.
A transcript of the meeting is being kept, so it's requested
that speakers identify themselves, speak clearly and plainly and into
the microphone, so that the transcripter can get you on tape.
This promises to be a very exciting meeting to me. We have
some very distinguished people here.
We have the staff, who's willing to come and share some of
their views with us, and we have three invited experts with us this
morning, all of them former office directors of the Nuclear Regulatory
Commission and now highly-regarded consultants.
Our three invited experts are Bob Bernero, Bob Budnitz, and
I have some introductory comments that talk about these
people. I guess I'll just read them.
Mr. Bernero spent 13 years in naval and space nuclear work
at GE and then served for 23 years, from 1972 to 1995, with the AEC and
NRC regulatory staff.
After five years in reactor and fuel cycle licensing, Bob
began work in regulatory development, including decommissioning
standards and spent fuel licensing.
After investigating the TMI accident, Bob formed the
Division of Risk Analysis in the Office of Research, served later in NRR
licensing divisions, and then went back to NMSS until he retired as
director in 1995.
Dr. Budnitz worked at the University of California Lawrence
Berkeley laboratory from '67 to '78 and held the position of associate
director and head of the Energy and Environmental Division.
In 1978, the joined the Nuclear Regulatory Commission as
Deputy Director of the Office of Research and was appointed Director of
that office in '79.
In 1980, Bob left the NRC to found the Future Resources
Associates, a consulting firm working mostly in risk analysis.
His current consulting activities include PRA, emphasizing
external hazards, upgrading the safety of older reactors, and using risk
in safety regulation, including performance analysis of waste disposal
Dr. Murley was the Director of NRC's Office of Nuclear
Reactor Regulation from 1987 to 1994. Prior to that, he was the
Regional Administrator of NRC's Region I office, beginning in 1983.
Dr. Murley retired from NRC in 1994 after 25 meritorious
years of service. He is presently a consultant on nuclear management
and safety matters in the U.S. and foreign countries.
In addition to all this brain power and good thoughts,
you're going to be treated early on this morning with some thoughts on
this subject from me and Dr. Garrick and from Dr. Apostolakis, and by
virtue of this awesome power I have as chairing this committee, I've
decided I'll go first and get things started and then turn it over to
John for his comments and then let George run the sprint lap and make up
for all the time we've overrun.
So, I do have view-graphs, so I'm going to do this and move
up to the front.
I am going to give you some thoughts I have on this subject,
to put it in somewhat of perspective. These thoughts are my own, by the
way, and may or may not represent any of the views of the ACRS or the
ACNW. For that matter, I don't even know what the ACRS views are on
this topic, or even if they have any.
So, they are my own.
That disclaimer said, I do have a couple of concerns that I
hope we can at least address in this meeting.
The first concern I have is there are a number of
definitions to defense-in-depth that vary slightly from one to the other
that I've seen.
Most of these definitions have a component of
defense-in-depth is there to compensate for uncertainties in our risk
Well, I think we can all agree on that, but the problem I
have with that is I can't use that. That's not enough. That's not a
definition. It's a sort of a description, and I have no way to
implement that in that regulations or to use it when I design some sort
of system to deal with the risk.
So, that's the first problem. I don't know how to design to
that, and we need a better definition.
The second problem is what definitions I have seen don't
lend themselves in any way that I can tell, except in an arbitrary
sense, of determining necessary and sufficiency conditions on
We've had a number of instances where there's been arbitrary
appeals to defense-in-depth to disallow some change or some regulation,
and if we're going to reap the benefits of risk-based or risk-informed
regulation, we have to have a way to put rational limits.
We have to know what defense-in-depth is, we have to be able
to identify it, and we have to be able to say how much of it is enough,
and I hope -- I don't think we'll resolve those two things at this
meeting, but I hope we at least make some headway in addressing it.
MR. APOSTOLAKIS: Tom?
MR. KRESS: Yes, sir.
MR. APOSTOLAKIS: I think language is extremely important
here. So, I would change a little bit something you said earlier.
You said "arbitrary appeals to defense-in-depth." The
appeals do not have to be arbitrary, because defense-in-depth itself is
MR. KRESS: Yes. Good point, George, and I agree with that.
As a way to approach the subject matter, I hope today we can
-- if you notice, in my title, I had the word "design" defense-in-depth.
I hope we can focus on that, as opposed to operational.
I don't want us to get sidetracked into things like
inspection, procedures, quality assurance, management, and even
While those things are considered components of
defense-in-depth, I think if we're going to address a true definition of
defense-in-depth that has ways to put limits on designing facilities to
deal with risk, we ought to focus on design aspects, and in addition to
that, we have a tendency to lapse into barriers and nuclear reactor
defense-in-depth as it's traditionally been covered or been looked at,
and I think we need to generalize the concept, generalize it in the
sense that it applies to any hazardous activity, and in order to do
that, I've put together what I call four design defense-in-depth
principles that I think are general and would apply to just any
The first one is do what you can to prevent accidents from
starting in the first place. That's, I call, initiation or paying
attention to initiating events.
Second is do what you can to stop accidents at very early
stages before they progress to unacceptable consequences. I call that
The third is do what you can to provide for mitigating the
release of the hazard vector. The hazard vector in nuclear power
reactors are the fission products, but it could be toxic gases or fire
and smoke or heat or whatever the hazard is you're dealing with. I call
that one mitigation.
And fourth, provide sufficient instrumentation to diagnose
the type and progress of any accident. Call that, of course, diagnosis.
And I've categories these, the first two, as prevention and,
with some overlap, the second and third one as mitigation and the fourth
one as belonging in both categories.
So, I've categorized defense-in-depth principles in terms of
prevention and mitigation.
Now, with those as sort of principles of defense-in-depth, I
think one could arrive at a definition of defense-in-depth, and I think
we may hear several of those today.
I have one that I prefer, so I'm going to propose it right
now, based on these kind of principles.
A generalized risk-related definition of defense-in-depth
could be -- and I'll just read it -- design defense-in-depth as a
strategy of providing design features to achieve acceptable risk, in
view of the uncertainties, by the appropriate allocation of the risk
reduction to both prevention and mitigation.
I like this definition for a number of reasons.
One, it, I think, captures the essence of what we
traditionally think of as defense-in-depth, and number two, it is linked
explicitly to risk analysis and risk concepts, and number three, I think
it lends itself to being able to provide limits to defense-in-depth, and
you may ask how can I work from this definition to arrive at limits?
Well, the key words are "appropriate allocation."
In order to arrive at limits on defense-in-depth with a
definition like this, first off, you do have to have risk and acceptance
criteria for the activity you're dealing with.
These are things like, in nuclear reactors, early death,
latent fatalities, land interdiction, could be frequency of fission
product release or could even be LERF as a surrogate for all of those,
but you have to have an overall risk acceptance criteria, and not only
that, you have to express these risk acceptance criteria in terms of the
If we're going to deal with uncertainty by defense-in-depth,
we have to have some quantification of what that uncertainty consists
Now, you may hear that there are two kinds of uncertainties,
those that you can quantify and those that you can't.
I maintain that if we're actually going to put limits on
defense-in-depth, you cannot have un-quantified uncertainties; you have
to quantify the whole thing.
What we normally call quantifiable uncertainties can come
right out of the PRA.
What we normally call un-quantifiable uncertainties, I
think, would have to have some estimate of what those are, and we'll
probably have to get that from expert opinion, for this
activity-specific and maybe even facility-specific activity, and the
acceptance criteria that I'm talking about in terms of uncertainties
have to include both of these.
Now, once you have that risk acceptance criteria, the next
question is you have to allocate it among those four areas of prevention
and mitigation, because that's what defense-in-depth basically is. It's
an allocation of risk. And how do you do that allocation?
Well, there's no differential equation or no technical basis
for doing it. Allocation is a matter of policy, and we have to have a
policy statement of some kind that says how much we value prevention
Now, that's policy, and I can't say how to do that, but we
could provide guidance.
For example, such an allocation or such a value judgement
could depend on the level of the inherent hazard. The more hazardous an
activity, the more we probably should value prevention.
It could depend on how big the uncertainties are. The more
uncertainty you have, you probably want to put equal balance on things.
It could depend on how much of this uncertainty is
un-quantifiable, as opposed to how much is quantifiable.
You may want to minimize the uncertainty. That would be a
classic optimization problem.
You might have noticed in my title I had "beating a dead
horse with a red herring." The dead horse is defense-in-depth as we
traditionally think of it. This minimization is what I threw in as a
red herring, just to confuse the issue.
It also -- some allocation rationally could be based on
what's called the loss function and decision theory. That's how one
normally allocates things. You ask yourself am I willing to suffer this
loss if I don't prevent? What are the consequences of that? And you
can work from that towards a probability that you want to accept for
With that as my introductory thoughts on the subject, I
guess I'll either ask if there are any questions or turn it over to John
Garrick for his thoughts.
I guess I confused everyone.
MR. BERNERO: Bob Bernero.
Are we going to reserve dialogue for the general discussion
period rather than take one paper at a time?
MR. KRESS: It probably would be a good idea to do it that
way. I think I prefer it that way.
MR. GARRICK: I think we're already in trouble
MR. BERNERO: So, I won't slap my forehead now.
MR. BUDNITZ: Bob Budnitz from Berkeley, California.
I have one very specific but, I think, important comment.
If you put a dangerous reactor 100 miles from the nearest
off-site person, then you have kept, as best I can tell from the
technology and what I understand it -- you've kept off-site fatalities
to zero, and that's a piece of defense-in-depth called siting and
mitigation, protective actions.
By the way, if you could do protective actions perfectly,
it's another piece, and you don't have that here. You only had the
piece about keeping the source term -- understanding it or keeping it
MR. KRESS: Bob, I agree.
MR. BUDNITZ: I think that's a crucial leg of this.
MR. KRESS: Yes, I agree with you that that is crucial
My reason for not discussing it, or even excluding it, was
there are lots of reactors out there that don't have that
characteristic, and we're talking about revising the regulations, and
we're talking about a lot of the NMSS activities in hospitals and
So, I was trying to say what would it be in terms of design?
I agree with you that that is a good defense-in-depth.
MR. BUDNITZ: But more to the point, if I have two identical
facilities that might be NMSS hospital licensees and one of them is in
the middle of nowhere and the other one's in the middle of New York
City, you might require different engineering at the facility, depending
on the site.
MR. KRESS: Probably not.
MR. BUDNITZ: You might.
MR. KRESS: Probably not.
MR. BUDNITZ: In principle, you could achieve the same
protection with different mixes of your allocation, but you don't even
know about that unless you put that allocation criterion on your slide,
which it wasn't.
So, I'm calling people's attention to the notion that you
have to consider that, I think, as a piece of this overall allocation
MR. KRESS: Yes. I don't know what all the criteria are for
allocation, I just know that we needed some, and those are good
John, you're up.
MR. GARRICK: I'm a little sorry I prepared anything,
because I would probably be more constructive if I took what Tom said
point by point and commented on it, but what I would like to do is come
before you not as a co-chairman of this meeting but as a plain vanilla
risk person and approach the problem from the point of view that, if I
had a license to do so, how would I address this question of
defense-in-depth, and again, as Tom said, I'm not speaking for ACNW or
ACRS, but I am trying to look at this as a issue that it's time that the
fuzziness of the issue was removed somewhat and that, in keeping with
the transition to a risk-informed way of thinking, it's time to think
about quantification of defense-in-depth as a way of taking the mystery
So, I looked at this from the standpoint of what might be a
conceptual framework for quantifying defense-in-depth, and I recognize
the various interpretations of what constitutes defense-in-depth from
the three fundamental lines of defense that have been articulated in the
material that we have received -- the plant, the safety systems, and the
consequence-limited systems -- as being somewhat of a classical display
of the three most talked about lines of defense, but even that can be
challenged, because there's the whole soft infrastructure of quality
control, of review, of assessment, of audit that people would argue very
strongly are and should be a part of defense-in-depth.
But the position I'm going to take is what we need to do is
pick a piece of it and start looking at it in terms of how we might
So, the piece that I have picked is to look at a reactor
example, have a license to do that as a risk assessor, and a waste
example, and one of the things, I think, that would help this process a
lot would just be to organize the way in which we talk about it and the
way in which we present it, and one of my favorite presentation formats
is a matrix format, a two-dimensional array, and if we have more than
two variables, I have a tendency to fix those variables in some fashion
and reduce it to a manageable presentation.
So, what I have chosen to do, to illustrate, at least
conceptually, what I'm talking about, is to look at protective systems,
protection systems, again admitting that defense-in-depth is more than
protection systems, but to take a very top-down perspective of it, and
having just spent three days on a safety committee at a boiling water
reactor in a very upbeat situation where it's a plant that had its best
all-time performance year, broke all kinds of records in terms of
capacity factors and availability, had the longest run of any plant, any
boiler in history between outages, received an INPO-1 certification, and
it's kind of exciting, and when I'm at the PWR, maybe I'll do the PWR
But what I'd like to do is to suggest that, if we laid out
the information about a reactor in some fashion similar to this, in a
top-down fashion, this is at the very functional level, and say that the
safety functions are basically those -- reactivity control, inventory
control, by which we mean coolant inventory control, heat removal -- as
we all know, the panic in Three Mile Island in the first two or three
days after the accident was a search for heat sinks, and radio-nuclide
containment, and then, in the vertical direction, we talk about classes
of initiating events, and I won't even claim that this is complete, but
the idea is to make it as complete as possible, and generally, we can
divide that into these three classes -- loss of coolant, transients, and
external events, and generally, we can create information that would
allow us to construct probability curves associated with those kinds of
events, and I think we could also argue that, in most large-scope PRAs,
we could aggregate the information in this form.
So, each of these kind of represent a group of scenarios,
and this is the end state core damage frequency for the group of
scenarios that are initiated by loss-of-coolant events, and then the
question -- and then, of course, if we do this carefully and we
probabilistically sum these end states, that constitutes our core damage
frequency, our total core damage frequency.
Now, the question is what do we put into these grid boxes,
and that's what I'd like to talk about a little bit, and I also would
like to reduce this from the very functional level down to a more
hardware level to give it more physical meaning.
Well, there's any one of a number of things and combinations
of things we could put in those grid boxes, but here's some suggestions.
Certainly, in each function, we could put the function
unavailability in terms of the frequency per demand for that class of
initiating events, and also, we could put something like this.
We could put what the core damage frequency would be at the
end state of that particular class of initiating events, given that that
function or system was unavailable. That's material that we can all
extract from a full-scope risk assessment, with some debate, of course,
but the most important entry might be this one. It's the total core
damage frequency with and without the safety function.
This particular core damage frequency is a result of the
convolution of all of the scenarios, and this is the same thing but
without the safety function, and at least if we did that for each of
these grid boxes, we would begin to see what the perspective was of the
contribution of the various safety functions.
Now, if we look at that at a slightly more detailed level
for something like a BWR -- and every time I look at this, I want to
re-tune the labels, and I'm not going to apologize for the small of the
print, you've got copies, but the safety functions can be reduced
basically into vessel-level make-ups systems and a reactor coolant
system, and the one thing you have to remember, that to a risk analyst,
we don't think in terms of safety-related and non-safety-related
Every system has to prove that it's non-safety-related.
So, I'm not adopting the classical NRC language here, but I
am adopting the classical risk language as to what these systems are
labeled and look like.
So, we have turned up the microscope on one grid box of that
functional diagram, and that's the grid box "inventory control," this
So, the figure I just showed you is just a blow-up of this
one versus this class of initiating event, and we've decomposed that
into eight safety systems and six categories of initiating events.
These are still categories of initiating events, and so,
when we talk about these entries, we're talking again about the total
core damage frequency being the probabilistic sum of the end states of
all these different categories, and then the curve that we want to
compare that with -- this should be a double curve -- is the curve that
results -- that comes about as a result of making the system of interest
unavailable, recalculating this end state, and adding that recalculated
end state to the rest of these and comparing that with this, gives us an
in-context perspective of what that system is providing us with respect
to the bottom line, and that seems to me one of the things we want to
Now, how do we do this with respect to nuclear waste?
Quite a different problem, because here passive systems
dominate the analysis, not only passive systems but geologic natural
setting are a major part of the analysis, and again, you can think of it
functionally, and I apologize to the performance assessment people for
choosing my own labels here, but I see the performance assessment
problem at the protective barrier functional level as basically these
three things -- water location and flow control, waste package
containment, and source term creation, mobilization, and transport --
and in a sense, you might look at this as the base case, and I have also
put down here geo-technical events to account for earthquakes, igneous
activity, and anything else of that type that you'd care to include, and
in principle, given the way we've set this up, the performance
parameter, in principle you could add these probabilistically.
Now, the way I've eliminated the time dependence of the dose
is to choose the time at which the annual release is the maximum into
the biosphere, and that allows me to keep it in a two-dimensional space,
and so, what this is is the peak annual release to the biosphere in
curies, and of course, this is just an expression of the uncertainty
about that, hopefully reflecting both information uncertainty and
Now, this time, however, what we want to do is, if we remove
this function, what does this curve become, and compare that risk curve,
which would be the one on the right, with what it is if you had the
function. In other words, this curve, the one on the left there, and
this one is the same, with all systems performing their intended
function. So, what that is is here.
This would be the measure of the performance with and
without the function or the system.
Now, how would we decompose that one, just to, again, reduce
it into more physical descriptive terms? This is how it might be
As far as water flow and spatial control systems, you could
imagine these kinds of systems, systems that would somehow impact the
way in which the water from the rainfall is drained from the site, and
I've distinguished between water diversion systems that are brought
about by doing some engineering of the geology versus bringing
engineering systems into the near field, and as far as waste packaging
-- and I'll let you argue as to whether things like drip shields would
be here or here. I would put them here.
Waste package containment -- I'm talking primarily about the
performance of the waste package, and usually there we think in terms of
the waste package corrosion resistance capability, fuel cladding, and
what have you.
Now, as far as the creation of a source term is concerned,
some of the things that are involved here are whether or not we have a
back-fill for purposes of enhancing geo-chemical conditions, also how
much credit we give to things like solubility, retardation, dilution,
and so forth.
So, again, it's a retaining of this structure such that you
have components where you can get some visibility into the contribution
to the overall performance of the taking away from, modifying, or
changing or adding any particular system/subsystem at any particular
So, I wanted to do this because I think that the hope here
is that we take advantage of what we've learned in the risk field.
I think most of the kind of calculations that we're talking
about here have been done.
We can debate about the quality of them, we can debate about
whether they contain the right kind of uncertainties, but that's okay.
Once we get it in this kind of form, and given that those
kind of issues apply to all the boxes, there is great value in the
comparisons, it seems to me.
So, I wanted to just throw this out as an opening salvo, and
as I say, we're in trouble on time, and the chairman, and particularly
me, have contributed to that, and we'll take questions but probably
MR. KRESS: The reason we're in trouble on time is this is a
particularly long-winded group.
MR. APOSTOLAKIS: I was asked to do two things: one, to
present some thoughts that Dana Powers had, the chairman of the ACRS,
and he couldn't be here to present them, and since I happen to disagree
with him on a lot of things, the committee felt that I was the best guy
to present his ideas, and then, I will present some of my own thoughts.
So, we start with Dana.
He gives us first -- and you have the write-up in front of
you, plus the view-graphs -- a sort of historical background on
This is a concept that has evolved over the years, from the
early days when people realized that there were -- there was a
possibility of catastrophic accidents from reactors, the uncertainties
were very large regarding the likelihood of occurrence, so people
devised this idea of multiple defenses.
It turns out, though, that this safety strategy that's
called defense-in-depth may impose unnecessary burden now on the
Everybody says that it has served the reactor safety
community well. I have some doubts about it, but I will go along with
Oh, I'm sorry, I'm presenting Dana's.
Even within the reactor safety community, thoughts have
turned to limiting defense-in-depth.
Now, you probably have seen that paper that several of us
wrote and presented at the PSA conference last August where we
identified two schools of thought.
One is the structural school of thought, defense-in-depth,
and Dana is the primary advocate of that, I believe, which says that,
essentially, defense-in-depth is an idea that is embedded in the
regulations, this idea of multiple defenses.
The rationalist school -- Tom and I happen to push that a
little bit -- advocates that defense-in-depth -- that now that we can
quantify uncertainties, we can use defense-in-depth in a more limited
way for those uncertainties that have not been quantified.
Dana offers a couple of thoughts here, says that the
structuralist approach may be difficult to extend in other areas -- he
has in mind NMSS activities, other than reactor, in other words --
whereas the rationalist approach could be extended to other areas, but
then, since you are relying so much on what can be quantified and what
cannot be quantified, you really have to have the analytical capability
which perhaps does not exist in other areas.
Now, a favorite question that Dana raises is what if you're
wrong? That's why I use defense-in-depth. What if my analysis is
So, he says that it may be a little paradoxical to use
analysis to specify where defense-in-depth is applied when, in fact,
defense-in-depth is used to protect you against the possibility that
your analysis wrong.
So, that's an interesting thought there.
So, again, some of the historical reasons for the
development of defense-in-depth here -- again, always according to Dana
-- at that time there was little experience in the operation of nuclear
power plants, there were no industrial standards for the safe operation
of nuclear reactors, there was confidence that accidents were unlikely
but great uncertainties in the consequences given that they would occur,
that they occurred, potentially consequential accidents would be
difficult to interdict once underway, and finally, if an accident
happened at one facility, it would affect the operation of other
facilities, as well.
So, Dana's conclusions are that, for the four classes of
NMSS activities, which are disposal of high-level waste, engineered
casks for transport of nuclear materials, sealed and unsealed sources --
I don't remember the third one, some sort of waste -- Dana believes that
the consequences for these classes of material licensees can be easily
In many cases, there is a wealth of operational experience.
I'm glad he said that, because I want to use it later.
The timing is different. Severe accidents are potentially
-- have large consequences develop slowly, so there is the possibility
to interdict, unlike with reactors. Phenomenological uncertainties are
modest, and the technical basis for rationally limiting defense-in-depth
is not well developed.
So, his main position is that he is against the imposition
of a defense-in-depth philosophy on material licensees, which I guess
includes high-level waste repositories.
Now I will present you my thoughts.
The fundamental question is why do we bother? Why are we
having this meeting? What is it that has changed over the years that
has made us have meetings like this, publish papers, and think about
defense-in-depth and its role in reactor regulation?
I believe most of us would agree that the thing that has
changed is that the uncertainties that forced the pioneers to come up
with defense-in-depth now -- a class of those uncertainties can be
quantified, whereas in those days they could not quantify them.
They knew that the frequencies of these accidents were very
uncertain, the consequences could be very high, but the uncertainty was
not quantifiable at the time.
In the last 25 years, starting with the pioneer in reactor
safety study, of course, we started quantifying a good part of these
uncertainties, and again, people with some experience in the field know
that there is also a class of uncertainties that perhaps we cannot
quantify at this time, un-quantified uncertainties.
The potential conflict, then, is between someone who takes
defense-in-depth as a principle and someone who tries to use the
rationalist approach and use defense-in-depth or its tools as standard
engineering tools used within engineering calculations that include risk
assessments and quantification of uncertainty.
So, what I propose is that we avoid the word "principle" and
simply say limit defense-in-depth and say defense-in-depth is a safety
philosophy that requires that a set of provisions be taken to manage
un-quantified -- not un-quantifiable -- un-quantified uncertainty
associated with the performance of engineered systems.
I believe this is consistent with Tom's presentation.
So, I'm carefully avoiding the word "principle." I'm using
the word "safety philosophy." In this, of course, "un-quantified
uncertainty" are the key words here.
Now, some observations.
Many times, people use the words "defense-in-depth" to mean
Now, by "barriers," by the way -- the word "barrier" is very
general here. It includes siting, it includes everything, not just
physical barriers like the primary system coolant boundary, and I want
to make that distinction. They are not identical concepts.
Even within the quantified uncertainties, where I'm going to
be using, you know, risk to decide how much I need, I will use multiple
barriers, otherwise I will never be able to go down to 10 to the minus 4
and 5 per year, but this is not using defense-in-depth; this is using
standard engineering tools.
So, let's start by saying that these two things are not the
Now, where does this un-quantified uncertainty come from?
It's primarily from models. We know that our models are inadequate in
many instances, or we know that some of the things that may be important
we cannot even quantify, we haven't tried. Okay?
So, experienced analysts and practitioners do have an idea
how good these analyses are.
Now, if we focus on these un-quantified uncertainties, then
we have to debate them, and then we will all understand better why these
uncertainties are not quantified.
We may be able to define new activities, research activities
or other kinds of activities, experiments, perhaps, to quantify part of
these uncertainties. So, it's not that I'm ignoring them. I think I'm
placing extra attention on these un-quantified uncertainties.
But the crucial question, as I said earlier, is under what
conditions, if any, is defense-in-depth in principle? I don't think
there are any conditions. It should never be called a principle.
It's a safety philosophy, as I gave in the definition, where
the uncertainty is un-quantified, and the words should not appear at all
within a PRA.
When the uncertainties are quantified, drop
defense-in-depth. You just use the tools to manage your risk and
achieve the uncertainty levels that Dr. Kress talked about.
Now, Dana read this and said, well, I am much more
comfortable with defense-in-depth as a means to address the question of
what if we are wrong in our analysis. This is his favorite question:
What if you're wrong?
You can argue that this is just a kind of uncertainty, as,
indeed, I am arguing, but I think that argument trivializes the problem
or implies that we know more than we do.
Well, instead of defending my position, I will attack his.
This is exactly what's wrong with calling it a principle.
You are telling me, no matter what you do, what if you are wrong? So, I
will impose on you defense-in-depth.
Well, I might as well give up. Why did we even try to
develop PRAs? We spent all these resources the last quarter-century.
Why? What if I'm wrong?
I will have to live with defense-in-depth forever, and
that's exactly what the word "principle" does to you. If you call it a
principle, you can't get out of it. It's impervious to analysis.
And in fact, I'm glad that he said, in his presentation --
it's really kind of unfair that he's not here, but on the other hand,
there is a certain pleasure in this.
Why is this a reason to argue against the imposition of
defense-in-depth on material licenses? Why? Because there are no
un-quantified uncertainties. That's why.
Thank you very much.
MR. MURLEY: My name is Tom Murley.
George, I very much, I guess, like your analysis her, but
are you suggesting that one should not make it a principle and,
therefore, if you are confident enough, you could use PRA to justify
removing a barrier like containment, let's say? Would you push it that
MR. APOSTOLAKIS: First of all, I would not use just PRA; I
would use my total knowledge. Yes, I would. Yes. There is nothing
sacred about the containment. But you better come back with some real
good physics to convince me that the uncertainties are not large.
MR. KRESS: George, if you adopted my principle of
allocation, you might say that allocating risk reduction to CDF and to
containment is a matter of policy, and then you would set values for
that allocation, and you would have a containment, even though you could
throw it away and still achieve your risk acceptance, you would still
have containment, because it's a policy in allocation.
MR. APOSTOLAKIS: On the other hand, you might say that the
policy applies to a certain type of reactors -- LWRs, for example.
If somebody comes up with a new design that is fundamentally
different and can make a convincing case that I don't need the
containment, I don't see why I should.
MR. KRESS: I guess now we turn to the rest of the agenda,
if I can find it.
That covers the preliminary presentations by the committee
members, and the second part of the agenda -- and we're only about 25
minutes behind, which is not too bad at all -- is presentations from our
invited experts, and we have first on the agenda Dr. Budnitz.
MR. BUDNITZ: Twenty years ago this week, I appeared before
the ACRS downtown, and Bob Bernero reminded me that I sat up on this
side of the table, with my jacket off, tie off, shoes off, and talked
this way, but I won't do that today, because Chet Siess isn't here. May
he continue to prosper. Those were the informal days.
I also want to point out that the reason I'm first is I'm
the youngest of the three, and another reason why I'm first is because I
was the Director of Research for a very brief micro-second 20 years ago,
and these two guys were two of my division directors.
I'm going to confine my remarks to Yucca Mountain and Part
63, but before I do that, I want to start with a bit of philosophy,
because I want to be sure you understand that I think the argument about
whether it's a principle or a criterion is moot, because it depends on
how it's used, and it's only how it's used that matters.
Let me try to make the point directly.
In Exodus, there are 10 commandments, and the two that, by
the way, are observed almost universally in all societies everywhere are
don't steal and don't murder. Don't steal and don't murder.
Are they requirements? Are they laws? Are they what?
I can tell you that, in the United States, in the year 2000,
we are still arguing about the definitions which goes to the
implementation. What really matters is the implementation of those
For example, we're still arguing today about whether
abortion is murder in this country. So, it's not simple just to say
Second, can I steal from my community property from my wife
in California? It turns out that's ambiguous. There's no real answer
to that in California law.
So, things as simple as don't steal and don't murder, which
are principles which all societies follow -- never minding they're in
the Bible, all societies follow them -- can't be implemented without
implementing rules, and it's the rules that govern our behavior, our
enforcement, our regulations, and not what you call it, whether it's a
principle or a biblical commandment or what.
Same thing is true here, and when you come to see what I'm
going to say about Yucca Mountain, you'll see it directly.
George, I don't know what to call it, but one thing for sure
is that, whatever you call it, at Yucca Mountain or for reactors or for
material licensees, it's -- what matters is how the rules and
regulations of Part 50 or Part 60 or Part 63 or whatever, or any of the
regulations, and all the stuff that goes with them, how it's used in
practice, and that's the real point.
In a way, you can imagine that they're high-level criteria
or high-level requirements which, if you meet this stuff, you meet it,
but you can't meet it by itself. You don't know how to meet it by
itself. You've got to meet this stuff that's down below, and then, by
definition, you meet it.
But using it as a principle, then, or a philosophy or
whatever, is because it provides a intellectual framework or a way of
thinking about how this stuff works or how you got to it, and you can
argue about it, you can argue about the details in light of those
principles which you think about, but you have to keep that in mind.
You can't enforce defense-in-depth anymore than you can
enforce what the Atomic Energy Act in 1954 ordered the AEC or NRC not to
do, which is to ensure adequate protection, but you can't go to any
licensee anywhere and say, sorry, you don't meet adequate protection.
What you say is you don't meet part something-something of Part 50.
That's what you don't meet.
By the way, that got translated later into no undue risk,
and it took the Commission 30 years to decide what undue -- you know, as
Hal Lewis on this committee used to say, you really want them to tell us
how much risk is due. That's the safety goal.
The safety goal finally told us, for reactors, what undue
risk meant, even though undue risk had been used for 30 years before the
safety goal was adopted.
You couldn't regulate on undue risk. You can't regulate on
adequate protection. What you can regulate to is some rule somewhere or
what an inspector is told to look for or what can be enforced, and
that's what I'm going to talk about for Yucca Mountain.
So, now I'm going to talk about the dilemma, and this is
quoting straight from the supplementary information for Part 63 that
came out within the last year, where it says in plain English, or
reading the plain English -- and then we're going to come to, you know,
where the rubber hits the road -- the Commission does not intend to
specify the numerical goals for the performance of individual barriers.
By the way, this is a draft; it still hasn't been finalized.
But were this adopted, it tells us the Commission does not intend to
regulate specific numerical goals for barriers.
But -- and here's the big "but" -- in implementing this
approach -- the defense-in-depth was in the previous sentence, so that
insert is, in fact, completely -- I'm not fooling you -- the Commission
proposing to incorporate flexibility into its regulations by requiring
DOE to demonstrate the repository comprises multiple barriers but does
not prescribe which barriers are important or describe their capability.
Don't steal -- but without telling you what stealing means.
I'm just reading the page. Okay? You can't implement don't murder or
don't steal without the details. You can't, because there are
ambiguities about what it means.
MR. GARRICK: Disagree.
MR. BUDNITZ: Okay.
So, what it says here is kind of odd. Propose to
incorporate flexibility by requiring barriers, not going to prescribe.
Well, of course, they go further. So, it's not quite that bad.
This is just the next, you know, eight lines down.
The proposed requirements will provide for a system of
multiple barriers to ensure defense-in-depth and increase confidence.
Probably what you meant was so that you could increase
confidence, but I'm just reading it, what it says. I mean I'll give you
the benefit of the doubt on how you read it. Increase confidence so
that the objective will be achieved. Okay?
I just have to read it that way.
Now, here's the dilemma.
NRC, NMSS, Part 63, Yucca Mountain -- be sure you understand
Will NRC use this as a decision criteria? Which is really,
more directly, can DOE's license application flunk based on insufficient
defense-in-depth even if it would otherwise pass?
That's where the rubber hits the road, and then you've got
to get into some details about that, but that's the question, and it's
apparently yes. Of course, the rules aren't finalized yet, Part 63 is
still draft, and EPA has to come in and it has to get changed, but
apparently, yes. I've been reading testimony and talks and various
positions of the staff, and apparently, yes.
Now, if so, how? How will the decision be framed and made?
That's where we need to talk.
Observation -- and this is a crucial observation of mine:
The decision criteria, whatever they will be, need to be clear, they
need to be fair, and they need to be technically logical.
MR. KRESS: In other words, the Commission needs to revisit
this statement that they do not intend to specify numerical goals for
the performance of individual barriers.
MR. BUDNITZ: I'm going to argue that what's there is
MR. KRESS: Yes.
MR. BUDNITZ: And what piece of it they revisit, I'm not
sure, but what's there is ambiguous, and I know that the staff agrees,
because I've heard the staff say this in public, that more is needed,
and there are even some tentative positions, and I'm thrilled that
MR. KRESS: I think somebody needs to specify what those
goals for individual barriers are.
MR. BUDNITZ: Fair enough.
Now, I'm going to switch the order of my slides if you've
got them in front of you, because I'm going to make an observation.
I sent a letter to the docket on June 25, and I also sent a
letter to John Garrick, chairman of the ACNW, but this quote is from
both of them. This is from a letter that I wrote six months ago, seven
months ago. I'll read it to you, but you can read it, too.
When I apply these ideas to Yucca Mountain, I stumble
principally because the notion of so-called independent barriers, one of
which can fail without compromising the overall system, which notion has
been so useful conceptually for achieving and demonstrating power
reactor safety seems not to apply to Yucca Mountain, and everybody that
deals with Yucca Mountain understands this.
As I understand the design concept, one cannot assume total
failure of any of the so-called barriers without seriously compromising
overall performance, and that's not necessarily true, by the way, for a
I can show you power reactors operating in the world for
which, if you didn't have a containment, you could meet all the goals,
safety goals and everything.
MR. APOSTOLAKIS: I'm confused by that.
MR. GARRICK: Just one question. Where in a power reactor
does it say how much liquid control has to contribute to the risk?
MR. BUDNITZ: I understand that. I exactly understand, but
the idea here is -- without arguing about what works for reactors, the
idea here is that, for sure, you can't totally remove -- by the way, the
staff agrees with this -- you can't totally remove barrier number four
or barrier number one and still show it at Yucca Mountain, because it
doesn't work that way.
It's not the same as the fact that, at many power reactors,
you can totally remove the containment and you can still meet all
operating NRC goals, except the goal that says you've got to have a
containment, but you know, the overall safety goals and all that stuff
-- you can meet it.
MR. APOSTOLAKIS: I'm confused by that. This is a question
MR. BUDNITZ: Yes.
MR. APOSTOLAKIS: Are you talking about a particular
MR. BUDNITZ: I'm talking about a particular design.
MR. APOSTOLAKIS: PWRs as we know them today.
MR. BUDNITZ: Yes.
MR. APOSTOLAKIS: You are saying that, if I remove the
containment, I am not compromising overall performance?
MR. BUDNITZ: I am saying to you I believe that you can
still meet the overall safety goals for some designs.
Now, without arguing whether that's true or not -- I don't
want to argue that. What I'm saying is it is surely true at Yucca
Mountain that you can't remove -- totally remove -- and the staff is not
talking about that. That's what we're going to come to.
You certainly can't remove the canister. You can't remove
the ground. So, we have to talk about what I'm going to come to in the
next slide, under-performance, rather than removal, and that's where the
details come in.
Let's not argue about what I said here about reactors. I'm
talking about Yucca Mountain.
Now, I'm going to go back and say, in practice, perhaps --
and I don't know, I'm guessing. Perhaps in practice, despite NRC's
words to the contrary, DOE will never actually flunk at Yucca Mountain,
but defense-in-depth will be used, instead, like ALARA.
Do what you can beyond meeting the thing -- you met the dose
in Amargosa -- do what you can beyond meeting the bare regulations
whenever it's cost-effective or whatever you mean by effective -- again,
some other parameter that you have to pay for.
I don't know that, but that's one possibility as to how it
will actually be used.
But if that's true, how does NRC conceive this would work in
I mean there's the classic. Might NRC ask for protection
from one or another barrier in the name of defense in depth even if the
overall performance is okay? In other words, you met it, but you still
got to have a containment.
I'm not arguing this is bad. I just want clarity. You need
clarity, just as when you say don't murder, you need clarity whether
abortion is murder or not.
And then there's the classic: What if one barrier provides
90 percent of the total protection? Maybe that's not enough of a mix.
Go read the Congressional legislation, which says you've got to have
multiple barriers. But what if one of them produces 90 percent of the
Maybe DOE can say, great, we can weaken that barrier so it
only produces 40 percent and we still meet the rules. None of us want
that. That's nuts.
MR. GARRICK: Bob, you're missing, I think, an extremely
fundamental point that the pioneers had the foresight to put in the
fundamental Atomic Energy Act, and that is the word "reasonable."
MR. BUDNITZ: Oh, no, I understand, of course.
MR. GARRICK: And I just think this is nonsense, these
arguments, because they're not reasonable.
MR. BUDNITZ: Exactly. But that's why we need specific
criteria so that people won't use unreasonable arguments one way or the
other, without specific criteria.
MR. GARRICK: You don't have specific criteria in an of the
reactor -- Part 50 -- along the lines that you're talking about.
MR. BUDNITZ: Yes, we do. We tell you what the containment
must do. We prescribe its performance.
MR. GARRICK: You don't prescribe the performance of the
safety injection systems.
MR. BUDNITZ: No. We prescribe the performance of the
MR. GARRICK: I think you're splitting hairs.
MR. BUDNITZ: Let me go on. The staff has gone further than
this, thank God, because if you didn't go further than this, we really
would be in the soup, and that's what I'm trying to say.
You can't have don't steal up here. You've got to have some
detail that they have to meet or they don't meet or they can analyze
against, that you can regulate against, that you can decide, and the
designers can use, and so on.
If all you had was the dose in Amargosa Valley, you know,
dose rate per year in Amargosa Valley, and that stuff, the designers
know what to do. They know what to do. But if they've got to do this,
too, the NRC has the obligation to tell them what to do, tell them what
they're going to test against, what the criteria will be. That's what
So, we're talking here about under-performance. That's a
phrase I've seen recently. So, perhaps the staff isn't thinking about
-- don't assume total failure.
We all know that's nonsense. I don't know what total
failure means. What do you mean, total failure? We're not saying the
can isn't there. The can might not behave as well. We're not saying
the earth isn't there. We're saying maybe we didn't understand travel
times or maybe the chemistry is different than we thought. It's at the
extremes of some state of knowledge uncertainty distribution unluckily,
even though we think it's over here but we think it's possibly over
there, but maybe it really is over there.
So, maybe we're talking about under-performance rather than
-- you know, to assume under-performance of barrier number two or
whatever and go analyze it again. Fine.
What does this mean? And that's the point. What does this
mean? What analysis requirements leading to some sort of decision
criterion will satisfy my three figures of merit? It has to be clear
and it has to be fair and it has to be logical, and I haven't seen that
yet, and short of seeing that, to be argued about amongst the technical
community and understood, short of seeing that, you still haven't told
the Yucca Mountain project what they should do in their design and in
their analysis so that they know where they're going, short of that.
You need that. You need to have the details.
Now, finally -- this is really a place where I am truly
stuck -- if NRC lets DOE decide what under-performance means -- and
there has been talk about in some of what I've seen -- if DOE decides
that under-performance means this and says bring me the rock, wrong
rock, late in the game -- remember, they're designing it now, they're
finalizing their design now, and then they're going to analyze for a
couple of years, and that's a terrible dilemma. You just don't want
DOE will not assume so much under-performance that it will
flunk if, of course, it passes under the base case, you see, because
anybody can dream up a set of under-performances that will flunk.
I can do that, but in fact, isn't that just what NRC's
concern really is?
NRC ought to be concerned, as the regulator, in its
statutory role, to be sure -- they've got to look for combinations of
under-performance that might lead to serious compromises, whatever that
means, find out whether there -- what the probability is and the
consequences of those or how much we don't know or what the
uncertainties are or where we have to go get more knowledge and make
sure that's straight. That's NRC's regulatory job, as I see it, under
the philosophy of an independent regulator, right?
So, you just shouldn't ask DOE unless you ask them to
explore the whole face base, and then I don't quite know what to do with
that, because then it's the bring-me-the-rock thing.
So, perhaps NRC has to tell them how much to assume, and
that leads to the other problem, which I know the staff is wrestling
with, because I've seen discussions and so on, mainly NRC is trying not
to be overly prescriptive -- thank God, by the way -- in using the
philosophy of performance-based analysis and decision-making and so on.
So, this is the dilemma for defense-in-depth. The Yucca
Mountain project and the Department of Energy deserve specificity as
they're finalizing the design and doing the analysis.
MR. APOSTOLAKIS: What's under-performance again? I missed
MR. BUDNITZ: Under-performance is the assumption that
barrier number two or whatever, instead of totally fails, only fails in
a certain way. Just as we say, in the reactor game, analyze as if you
had a loss of off-site power, even if the probability is low.
MR. APOSTOLAKIS: But why do I have to tell DOE how much
under-performance to assume? Aren't they going to do it as part of the
MR. BUDNITZ: Well, the face base is so vast.
MR. APOSTOLAKIS: But they have to do this, assign
probabilities to these things.
MR. BUDNITZ: No.
As I understand it, they are supposed to produce a base-case
performance assessment, with its uncertainties explored, but they don't
necessarily have to show what the dose in Amargosa Valley is if barrier
number two under-performs by X percent or fails at 1,000 years instead
of 10,000 or has more juvenile failures than they think is right or
MR. APOSTOLAKIS: But if they assign probabilities to these
various scenarios, 1,000 years versus 5,000 years, then the performance
assessment will reflect all these.
MR. BUDNITZ: Only if they're asked to reveal it and if
they're told that that will be the thing against which they'll regulate,
Let me just describe a possibility.
Suppose I said to you that the department believes that
juvenile failures of the canister will compromise X percent -- it might
be X-tenths of a percent -- of all the cans. That's their best
estimate, and they have a uncertainty distribution about that state of
NRC might say I don't care what you do with that. Put that
in the performance assessment, but I want to see an analysis that's
In other words, instead of .02 percent, maybe 2 percent, as
a means of assuring that, gee, you know, I really don't know whether I
trust -- that's Dana Powers' argument.
That's a valid way to regulate, is to tell the licensee to
assume something that is unrealistically conservative and still show
you're okay, and that's not in the performance assessment.
MR. APOSTOLAKIS: Let me take an example of a PRA. Maybe I
misunderstand what you are saying.
Somebody brings me a PRA and I review it. That licensee
wants to use it in their process.
MR. BUDNITZ: Right.
MR. APOSTOLAKIS: The licensee cannot come to me and say I'm
not going to worry about common-cause failures, because you didn't --
MR. BUDNITZ: No. Let me make a postulate here that the
licensee, the applicant says we think that there are going to be five
juvenile failures of our canister in the first 5,000 years, and our
state of knowledge is such that we're very confident it's no more than
It's not inappropriate for the regulator to say analyze for
400 and show me what that does, and if you still perform -- that's not
inappropriate. If you still perform, great. On that aspect, we're
going to give you your license.
MR. APOSTOLAKIS: This is not a performance assessment
MR. BUDNITZ: We're regulating, George. That's just the
point. We're regulating. We're trying to regulate.
MR. APOSTOLAKIS: I'm playing devil's advocate.
MR. BUDNITZ: Of course. I understand.
MR. APOSTOLAKIS: So, DOE, the applicant, would like the
benefits of both performance-based regulation and the --
MR. BUDNITZ: No, no, no. Quite the opposite.
I can't speak for them, but they're probably thrilled with
just the single figure of the dose in Amargosa Valley, but if NRC is
going to say we're going to impose defense-in-depth by telling us that
we have to under-perform barrier number two as a means of exploring how
defense-in-depth actually works, somebody needs to write down what
under-performance means in detail so we'll know what to analyze, and the
under-performance is presumably outside of the realm --
MR. APOSTOLAKIS: You're really coming back to Tom Kress'
point that you have to have some sort of allocation.
MR. BUDNITZ: I'm not arguing that under-performance is the
way to go, but if they're going to do it that way, they need to
prescribe it, and it may be outside of the realm that DOE believes is
the real world, just as we said 2,300 degrees for the peak clad
temperature -- nobody thinks that's the right number, but if you meet
it, you get your license, and I'm worried that, absent -- and this is
early, soon, not five or 10 years from now -- I'm worried that, absent
specific criteria against which the department, Yucca Mountain, the
applicant can analyze and know that he passed or he didn't pass and can
change the design now, before it's too late, in order to, you know,
improve and meet, that it's an open-ended, unsatisfactory regulatory
MR. GARRICK: Bob, you seem to be strongly advocating an
MR. BUDNITZ: No.
MR. GARRICK: Well, you seem to be.
MR. BUDNITZ: No, no, no. I don't think defense-in-depth is
necessarily the principle that others do, but if they want to use it,
they've got to tell them how.
MR. GARRICK: The NRC has been very clear in telling them
that they want to know the role of the specific protection barriers, and
my whole point was that the only place that makes any sense is in
relationship to the bottom line.
MR. BUDNITZ: I quite agree.
MR. GARRICK: I think one of the things that's a problem
here is that -- the great thing about the PRA business is that we
established a measuring process through the PRA, and we got some
experience on it before we started fussing around too much and trying to
calibrate that measure, and I kind of see that here.
There are some fundamental principles that have been laid
down, and one of those principles is that all of the protection should
not come from just the engineered systems or just the natural setting.
MR. BUDNITZ: Sure.
MR. GARRICK: Now, it sounds like what you're saying is
that, if they say that, they need to say more about how much of it
should come from where.
MR. BUDNITZ: No, not necessarily how much of it should come
from where. I don't like that either.
They need to establish specific performance criteria or
analyses or outcomes or something like that that the department can
analyze to now, while they're still changing the design. Otherwise they
get the bring-me-the-rock problem.
They need to say under-performance of the canister means X
for juvenile failures, means Y for corrosion, means Z for when it will
happen, 1,000 years or 6,000 years. They need to tell them what
under-performance specifically means for those things, assuming, I
assume, that the under-performance they're going to tell them about is
outside of where the department believes is the true knowledge of the
Now, you know, I don't care whether you say it's this.
Analyze that anyway. That's not an illegitimate thing for regulators to
do, and they do that all the time.
MR. APOSTOLAKIS: Why would the NRC ask them to do that?
MR. BUDNITZ: Why don't you ask the NRC? But they're
talking about asking them to analyze under-performance of various of the
barriers, either one at a time or maybe in combinations, but absent
specific things, the applicant doesn't know what to do.
MR. APOSTOLAKIS: Are they doing sensitivity studies, then?
MR. BUDNITZ: Why don't you ask them?
MR. APOSTOLAKIS: It looks like you're saying the department
will come in here with a base case and what they think is likely and
this and that.
MR. BUDNITZ: Yes, sir, of course.
MR. APOSTOLAKIS: And then the NRC staff comes back and says
now do this, I would like you to do this, which is a sensitivity study.
MR. BUDNITZ: That's what I said. These are sensitivity
studies. They're always a good idea.
MR. APOSTOLAKIS: And is it because we feel that the
uncertainties -- that right now we cannot quantify them?
MR. BUDNITZ: Well, why don't you ask them? But here's what
I think, and I'm reading minds.
Apparently, somebody somewhere in this Commission and its
staff thinks that defense-in-depth needs to be invoked separate from the
TSPA, the performance assessment as a whole, taken with its state of
knowledge, and I'm not going to argue whether that's a good or a bad
philosophy, but if they want to do that, they need to tell the
department specifically, with specificity, what the things are to which
they're going to regulate, so they can change the design and show it's
MR. APOSTOLAKIS: I'm not familiar with that particular
staff position, but if, indeed, they want to apply defense-in-depth
independently of the PA, then that's exactly what I'm against, and I
hope I learn more about it.
MR. KRESS: In fact, it sounds like a de facto way of
Bob, did you have a question?
MR. BERNERO: Bob Bernero.
I'd just like to add -- I was going to address it in my talk
-- there is a statutory difference here.
MR. BUDNITZ: Yes, there is.
MR. BERNERO: The 11th commandment, not out of the Book of
Exodus but out of the Nuclear Waste Policy Act, simply says the
repository must have multiple barriers. So, there is a regulatory need
to address how does one implement that commandment, and that's part of
MR. BUDNITZ: Absolutely, but of course there's an easy way
to meet that.
The fact that there is engineered barrier design and the
earth is, by definition, a multiple barrier. If you really wanted to be
sloppy, you could say of course we've got that.
But if you want to go further -- and I agree with you, Bob
-- if the Congress wants to go further, got to go further, they've got
to go further specifically. They just can't let the applicant figure it
MR. APOSTOLAKIS: The words "multiple barrier" are so fuzzy.
Anything is a multiple barrier.
MR. BUDNITZ: George, the statute has that language, though.
MR. APOSTOLAKIS: Well, then it must be right.
MR. KRESS: I think we have time for more discussion later.
MR. BUDNITZ: Without specificity, it's like don't murder.
Without specificity, you don't know how to regulate.
MR. APOSTOLAKIS: I find that very interesting, Bob, because
in reactors we see the same thing. People want the performance-based
regulation, and you give it to them, they come back and say, what, you
didn't tell me what you want me to do.
MR. KRESS: Okay.
MR. KRESS: Will the meeting please come to order?
Now we're at the point on the agenda where we're going to
hear from Tom Murley.
You're up, Tom.
MR. MURLEY: Thank you, Tom and John. Thank you for the
invitation, also. I don't have view-graphs or slides, so I'll just sit
here and say my piece.
I should say at the outset that I am not sure just how much
I can help on your discussion on Yucca Mountain.
I've not kept current with all the latest policy statements
and SECY papers and ACRS letters and things, although I should say Jack
Sorenson did an excellent job, I think, in research this topic and
sending the material out, but I have given a good deal of thought over
the years to nuclear safety and defense-in-depth, and so, perhaps I can
discuss some philosophical issues, and if it helps you, fine.
The first point I guess I would like to make is that, in my
experience, defense-in-depth is not a regulatory requirement. It's not
a principle. It never was.
I would characterize defense-in-depth as an after-the-fact
explanation to Congress and to the public of how NRC achieves safety for
That is, after regulations were developed and after the
staff implemented them through branch technical positions and reg guides
and things, there was an explanation of what it all meant, and one way
to do that -- and I think a very useful concept -- was the
defense-in-depth concept, and as I read Cliff Beck's 1967 explanation to
Congress, that's probably one of the early things I read when I joined
the AEC in 1968, but it was never used as something that the staff used
as a requirement, a hard-and-fast requirement, and I think -- I'll give
This was illustrated by the Three Mile Island 2 accident.
I recall a meeting some months after the accident where an
aerospace safety expert was giving his views of the accident.
He may have been from NASA, and I think he might have even
have been assisting the Kemeny Commission, and he observed that NRC
talks about defense-in-depth but they don't really enforce it, and he
said, for example, the plant was designed -- this particular plant,
Three Mile Island, was designed for the pressurizer relief valve to open
during a feedwater transient so that the high-quality primary system was
deliberately breached during a design basis transient, and of course, we
know that the relief valve stuck open in that case.
He continued by noting that the operators defeated the
safety systems by shutting off the ECCS, the high-pressure injection,
and his point was that one of the major fundamental barriers of
defense-in-depth was deliberately defeated by the operator action.
We now know, of course, that there were confusing indicators
and circumstances that led the operators to take those actions, and
finally, this observer noted that the containment was open during the
early part of the accident and that that fact permitted radioactivity to
be released directly to the auxiliary building and to the atmosphere.
Eventually, of course, the sump pumps were secured and the
containment was isolated in that accident, but his point was this
philosophy of defense-in-depth was something that the agency, back then,
at least, talked about but didn't really enforce, and it was not -- his
point was, of course, a negative point with regard to the NRC and the
staff, and this analysis -- I'm sitting there listening to it, and I
became very embarrassed as a NRC staff member, because he was right, and
it had a profound impact on my thinking about safety at the time, and
that was, if NRC has a regulatory requirement and one relies on that
requirement in this defense-in-depth argument, then you really have to
So, you've got to make sure that the containment is reliable
and so forth.
In other words, the barriers of each level of
defense-in-depth should be highly reliable. That's the message I took
from that discussion, and it did follow me, and I did use it and think
about it during my career, at least, in that term.
I sent the committees -- actually, to John Larkins -- an old
document dated April of 1989 on Shoreham emergency preparedness that I
had in my files, and insofar as that was what we relied on -- that's
what I relied on when I licensed Shoreham in 1989, and it is, thus,
official Commission policy as of 1989.
So, it is a discussion of how emergency preparedness fits
into the defense-in-depth safety philosophy, and so, there's an
introduction in the first page of where emergency preparedness fits in,
and we termed it, then, as effectively a fourth level of safety. I
think that's the phrase we used.
Now, the significance of that paper for this discussion, I
think, is that the topic of defense-in-depth was used only as a
philosophical introduction. It doesn't say that it's a requirement.
I then stopped the discussion of where it fits in and went
through a point-by-point discussion of how Shoreham met the actual
regulations, and so, there was never a use of defense-in-depth as a
requirement per se.
As I said, it's kind of an after-the-fact explanation of how
NRC achieves safety, and my explanation -- I should say the agency's
explanation then, at that time, was that emergency preparedness was, in
effect, a fourth level of safety, but it was not meant to be that it was
an absolute barrier, or there were no numerical guidelines or
requirements for each of those levels.
There were other instances where I recall falling back on
the defense-in-depth philosophy in my own thinking about specific safety
issues, and I'll give a couple of examples.
The staff -- and I'll speak for myself, because I can't
speak for the staff today, but I was always sensitive to conditions or
accident sequences that could breach multiple levels of defense-in-depth
through a common cause, and we always paid a lot of attention to those.
That's why steam generator tube integrity was always such an
important issue for the staff. We gave it high attention, because
multiple steam generator tube ruptures could lead to bypassing
containment either before or after core damage, and that -- one may
wonder why, I guess, steam generator tube -- maybe it's obvious, but it
was for that reason, at least in my own thinking, that this was a path
that could breach multiple barriers of defense-in-depth.
And then in the late 1980s, I recall thinking about safety
culture and what does it mean, where does it fit into the overall
picture of safety, and it slowly became clear that and I concluded that
it was extremely important, safety culture was extremely important,
because -- it was Chernobyl, actually, that showed that a poor safety
culture at a plant could lead to actions that could cut through all
levels of defense-in-depth.
In other words, it could be a common cause for breaching
multiple safety barriers. If you've got a poor culture, you can do
stupid things that initiate the accident. You can do a test that's not
properly planned. You can put the reactor in conditions it was never
designed for. You can shut off safety systems.
In other words, it is a means for slicing through the
defense-in-depth barriers, and it was that thinking that personally I
went through that caused me to conclude that safety culture was an
extremely important safety concept. To me, it's not an abstract concept
or idea, but it's an essential aspect of nuclear safety.
So, I hope I'm giving some examples of how one regulator, at
least, on the staff used and thought about defense-in-depth.
There are some questions that were posed in the material
that was handed out to us, and I know Bob Budnitz and Bob Bernero have
talked about some of them, and I'll aim at a couple that I think I can
One is, is there an over-arching philosophy of
defense-in-depth, or a discussion of it, and I have not spent a lot of
time on the definitions.
I know there are lots of them, but the philosophy, to my
mind, is fairly simple, and that is, there should be multiple barriers
for protecting public from radiation, such that single mistakes and
single failures, even of programs -- like emergency preparedness is
really a program, you can think of it, but in that sense, as George
said, it's a barrier.
It doesn't have to be a physical barrier, and insofar as
possible, these barriers should be independent, and I don't think that
should be an absolute requirement, but one should try to make them as
independent as possible. So, multiple independent barriers for
protecting the public from radiation.
It should be made a regulatory requirement, in my judgement,
but it should remain a guiding principle, because it is a good way to
think about safety, as I think I've tried to illustrate.
A second question, how is it used in materials -- and I'll
let Bob Bernero, who's thought about this a lot more than I have and
also speaks about it better -- give some examples, but there's one that
I've come across recently that seems to me a perfect example of how
defense-in-depth thinking is used, and that is in criticality safety.
There is this concept of single contingencies, multiple --
double contingencies, triple contingencies as protection against
criticality, and that, to my mind, is a perfect illustration of how one
thinks about multiple barriers of defense-in-depth.
Apparently there is -- well, I know there is a lot of
discussion of how should PRA be used in risk-informed regulation
consistent with defense-in-depth, what does that mean, and I guess I
don't have the answer to that, but I can tell you how I interpret it,
and that is it means don't use risk arguments solely to weaken or remove
levels of defense-in-depth.
I think that's how I would use it if I had to use that
language, and even though one has to, I guess, hold open the theoretical
possibility, George, that you could use risk arguments or numerical
arguments to remove containment, that comes very close -- well, it's a
regulatory requirement, so you probably can't do it, but it comes very
close, I think, to using defense-in-depth as close to a requirement.
MR. APOSTOLAKIS: I'm coming back to Bob's question of what
is murder? What is a risk argument? A risk argument, in my view,
includes all the engineering analysis and physics that is appropriate to
So, in my mind, one could use risk arguments to reduce
defense-in-depth, as long as the uncertainties are handled properly and
So, a risk argument -- I mean PRA, in my mind, includes the
underlying physics, chemistry, and engineering that sometimes we call
So, I assume that's what you mean by risk argument?
MR. MURLEY: Yes. And I did not say and I certainly didn't
mean to imply that you cannot use risk arguments or engineering
analysis, the whole panoply of arguments, to reduce margins where
they're excessive and that sort of thing, but I think you would run
across some severe resistance if you pushed the argument to remove an
entire barrier of what people view as defense-in-depth.
For example, people have used the argument, risk arguments
-- and I've heard them -- to remove emergency planning, period, for
advanced reactors. I think that's going to run into some serious
programmatic, you know, policy problems.
I think it can be used to quantify the protection offered by
these levels, and I think John Garrick's paper -- I did skim it, and I
did listen to him carefully. I think it's a very good analysis, an
appropriate use of how to analyze and understand barriers.
If it's pushed to the level of using numerical goals for
those barriers, then I think that's maybe pushing things a little
further than people are ready for today, although in principle, one has
to hold open the possibility that it can be done.
There is the notion of safety goals. Are they clear for
regulatory use in the materials area or even the reactor area, for that
matter, and I must say the safety goals -- I found them to be not much
use at all.
The public health goals -- I'm sure you realize, of course,
there's a big gap -- there's an order of -- two orders of magnitude
difference between the public health goals and the plant performance
goals in terms of the protection that they offer to the public, and this
has always been a stumbling block for use by the staff.
The staff was told by the Commission -- they worked with the
ACRS for years to try to rationalize a large early release goal with the
public health goals, and it couldn't be done, because there's this
One can have a TMI-2 meltdown accident every year and still
meet the public health goals. You can work it out.
So, they were not very useful at all, and certainly, when I
was with the staff, we didn't use them in our day-to-day activities,
with one exception.
We found them -- we did -- in reviewing and certifying the
evolutionary advanced reactors, we used a conditional containment
failure probability of .1 as a guideline, and we found that very useful
as a guideline, but even there, we had to back off using a numerical
goal, because -- in this case, it was General Electric complained -- and
I think they were right.
They complained that, in some cases, by forcing that goal,
you're actually increasing the core damage frequency.
So, we did is tried to formulate an equivalent deterministic
requirement that we felt was equivalent to the 10-percent conditional
containment failure probability, but overall, I have to say I don't
think that we found the safety goals very useful.
Finally, there is a nexus in all this discussion of
defense-in-depth to risk-informed regulation, and I'm a big fan of
I wrote a paper about it five years ago or so supporting it,
and I think I am very pleased with the way the agency is moving in this
direction, but there is a troubling aspect, and maybe I don't see it
correctly, but I would like to at least tell the committees what's
troubling me, and that is that there is a whiff in all of this
discussion, more than whiff, an aroma of relaxing regulations and
reducing burdens, almost as if this is a deregulation exercise, and you
know, there is room for that, I agree with that, but people forget the
other side of the coin, and that is there is this role of risk-informed
operation, too, where the operators of reactors, in particular, can use
risk to improve safety, and you can do them at the same time.
You can have reduced burden and improved safety at the same
time if it's done wisely, but I don't hear any discussion of that coming
out of this committee or coming out of the staff these days, or the
Commission, and I think somebody needs to pay attention to this, because
if risk-informed regulation comes to be seen as just a code word for
deregulation, I think the whole thing is doomed, because I don't think
you will have public support in the long run for that.
Some conclusions, then.
I agree with, I guess, John Garrick's characterization that
there is fuzziness in this defense-in-depth concept and that it can
stand some clarification and even some numerical clarification, and I
commend the committees for shining some light on this subject.
I am very uneasy with any notion of pushing defense-in-depth
to the level of a principle or a requirement, and I am also uneasy if
there is a trend to allocate numerical goals to the levels of
I think you'll run into trouble just like the safety goals
kind of ran into trouble, and ultimately, it would not be much use.
That concludes my remarks.
MR. KRESS: Thank you.
That brings us to Bob Bernero.
MR. BERNERO: I, too, would like to thank you for the
opportunity to speak to the joint subcommittee, and as I will explain in
my remarks, I'm going to try to focus more on the material licensing and
the high-level waste arena, or waste management arena, than on the
I would, however, like to start out with just an exposition
-- I used to tell people when I was here that the greatest conflict of
interest you'll face in your life is defending what you said yesterday,
and I feel a little bit of that now, because I'm going to go back to
statements I made in the past decades, when I was working in the NRC and
had the good fortune to be involved in safety goals and things like
that, regulatory philosophy.
A safety goal has practical use as a description of the
levels of safety or reliability that is sought by a regulatory system,
and similarly, a probabilistic risk assessment or any kind of risk
assessment has value as a description a display of your best knowledge
about the level of safety or reliability you are achieving but to
regulate to a safety goal, to define quantitative standards in a safety
goal as the formula for a safety decision on the acceptability of a
reactor or its features is not a wise move, and for years and years, as
safety goals were developed, there was a very strong philosophy that,
beware, don't regulate to safety goals, use safety goals in formulating
regulatory systems or approaches but don't regulate to the safety goal,
and of course, I will acknowledge that the high-level waste program,
from the very beginning, has as one, not the entire, but one basis of
acceptable judgement a safety goal.
That's what the performance assessment is calculating.
So, a word of caution on that, but talking here today about
defense-in-depth, as I will say shortly, defense-in-depth as an
approach, as a strategy for safety analysis, a strategy for design and
safety analysis, is a very good description of your caution in avoiding
undue reliance on any single feature, barrier, or thing or aspect, and
when you do that, your safety analysis should beware of a prescriptive
approach and the safety evaluation, with quantification where you can do
it, without quantification when necessary, or with very, very vague or
poor quantification, it still has to rely on reasoned judgement with the
best display of information before you and then make a decision.
Jack Sorenson gave us some questions. In the slides you
have, I slightly changed the questions, and I geared them so that I
could go through responses to the general questions and the specific
questions in the three specific areas of regulation, and that, of
course, would let me emphasize the ones I'm more familiar with.
I, too, would like to endorse the book -- I have it over
there -- that Jack compiled, the research on defense-in-depth. It's an
When I made the view-graphs, I consciously selected one of
the papers to quote from, and now I have forgotten which one, and I
don't think it's worth the research to go back, but the point is it's a
It's a good exposition not of a formula for adequate
protection but as a safety philosophy, and many of those definitions fit
Cliff Beck's 1967 one -- I was very familiar with that,
because I came to the NRC in reactor licensing, and that was treated
sort of like a gospel, but I think it was Tom or somebody said it was
more a public exposition of what we're about rather than a formula for a
licensee to build a reactor to.
Now, if I go to the very first question, is there an
over-arching philosophy, my answer is yes, there is an over-arching
philosophy as a strategy of safety analysis but not as a formula, and
the key thing here is the undue reliance on any single factor, a rarity
of occurrence, a design feature, a barrier, a performance model.
An example comes to mind.
Many years ago -- in fact, right now, it's more than 25
years -- I had the fortunate experience to be the licensing project
manager for TMI-1, and a principle safety issue and contention in the
hearing was adequate protection against the crash of a large aircraft,
because that plant sits not far from the end of the runway of the
Harrisburg International Airport.
There was a great deal of analysis to make sure that the
standard review plan, which was just developing at that time and used a
screening probability for screening out aircraft, that there was not
undue reliance on low probability of crash, and it ended up with a very
detailed analysis that included what would happen if an aircraft less
than 200,000 pounds hit, what would happen if the aircraft greater than
200,000 pounds hit, and one of the good aspects of it all was the
licensee, or applicant in this case, recognized all along that the
responsibility for developing a persuasive case to show no undue
reliance on that factor -- that licensee had that responsibility and
fulfilled it, and the staff didn't prescribe what was the due reliance.
The applicant demonstrated that there was not undue
Barriers are an issue peculiar to material licensing in many
Basically, as I've said, it's not a formula for defining
acceptability, and I would caution that simply because one has
defense-in-depth, that doesn't mean that there is acceptable safety.
You can have very frail defenses, and on those grounds, I
would suggest, when you move to the additional thought of risk-informed
regulation, that's going beyond defense-in-depth.
It is looking at barriers or dependencies or uncertainties
and seeking to achieve a sufficient margin of safety, not too much and
not too little, and it goes to the degree of knowledge that you can
have, or the degree of experience, in many cases, with material
MR. APOSTOLAKIS: Before you go on, Bob, I think one of the
issues before this subcommittee, I think, or maybe this meeting, is to
try to understand words like "undue reliance."
I'm trying to put it in the context of uncertainties.
Perhaps it would mean the same thing. When you say "undue reliance," I
would say I'm too uncertain about the effectiveness of these barriers
for some reason. Maybe I don't understand all the conditions under
which the barrier is supposed to function. I don't trust, perhaps, the
calculations that the event is really very rare and so on.
Would that be consistent with your thinking? Why is there
MR. BERNERO: Undue reliance -- as an example, in the TMI-2
case -- or TMI-1, actually. TMI-2 adopted the analysis verbatim.
In the TMI-1 licensing case, based on the traffic that the
Harrisburg International Airport supported and was reasonably expected
to support, a screening criterion like 10 to the minus 6, 10 to the
minus 7 per year likelihood of impact, using a conservative footprint
for the reactor plant -- that screening criterion was relied upon only
with respect to jumbo jets.
Basically, it was concluded that it is a relative rarity for
a jumbo jet, something substantially in excess of 200,000 pounds loaded
weight, to be in this airport or to be using this airport.
That left the screening criterion having (a) some good
traffic analysis as a basis and (b) the margin of safety implicit in the
robustness of the plant given that it was designed for aircraft up to
200,000 pounds, and it had things like a condensate storage tank on each
side of the reactor, so that your decay heat removal wasn't compromised
by the aircraft crash immediately.
You know, condensate storage tanks are out in the open. You
know, they're unshielded.
So, you had two things. You had an extraordinary
robustness, and frankly, the applicant said I'll change sites if I have
to get a degree of crash resistance beyond the inherent robustness of a
You know, a large dry containment is a very robust
structure, and they said that's what we'll do. We're willing to expand
this facility to that degree of robustness.
So the uncertainty of a screening criterion of probability
had two factors to make an evaluation: Is this undue reliance or not?
But there's no formula for that evaluation.
Now, our current safety goals and objectives -- I said a few
words about safety goals to begin with, but of course, it goes without
saying -- you're all aware that the current safety goals and objectives
are very explicitly reactor-oriented, and there's years and years of
that dialogue, and if you go into the material regulation or especially
into waste regulation, the only thing you find is in high-level waste
disposal the criteria that originally derived from the EPA standard, 40
CFR 191, which is a performance assessment with a quantitative release
limit probabilistically set.
So, I say they're not clear, because first of all, the scope
is not clear.
There's a span of protection or a scope of protection
implicit in NRC regulation that includes public safety.
In reactor regulation, you're almost always talking about
off-site public safety and not talking much about the worker safety.
That's within the NRC jurisdiction but not quite so
You know, look at the steam-line erosion/corrosion, that old
Surry incident, 1970-something, where a relief valve -- tail pipe came
out of the hole in the deck and scalded two workers to death.
Things like that -- NRC's jurisdiction for industrial safety
is not clear, and when you go into material regulation, you'll find that
ALARA for chronic exposure is an important aspect, but accidental safety
is dominated by chemical safety.
So, you have -- issues that are far more complex don't lend
themselves to formulation.
Go into medicine and there is serious challenge or question
about NRC's jurisdiction for patient safety -- you know, that is, the
person receiving nuclear medicine treatment, and of course,
environmental protection -- we have a congruence of NRC's
responsibilities and authority with EPA.
The practices at NRC, you're quite aware, has a very large
range, and I would just single out transportation, which I listed at the
bottom, as a very interesting example of lack of defense-in-depth.
Transportation relies on one barrier, a great big heavy,
bullet-proof, super-strong cask to hold spent fuel, and especially in
transport, you have one barrier, and the real question is not do I have
multiple barriers, but the real question is am I placing undue reliance
on that one barrier, and of course, here, you have a wealth of
experience, engineering, metallurgy, testing capability, quality
assurance. You have a variety of tools. But the test is, is there
undue reliance on a single factor or a single barrier?
Reactors -- I would just point out that, in reactor
technology, defense-in-depth discussions are, in my experience,
invariably associated with accidental releases, not chronic releases,
and that comes to be an important consideration in material regulation
and waste management, and of course, waste management is a chronic
The very nature of it is you take the waste and you put it
somewhere and say it will stay there until it's gone or forever.
In the reactor regulation area, seismic safety, here again
you have a probabilistic screen, and you have behind it -- some of you
certainly had an experience in the seismic margin analyses that were
popular a long time ago, and my favorite term, "HCLPF," the
high-confidence of the low probability of failure, which is a very good
concept, but it's interesting, if you ever go through the DOE
regulations and safety analyses for seismic safety, they actually try to
quantify, specific a specific requirement for seismic safety that you go
up to your design basis, probabilistically set, and then you go beyond
it by some formula and show that this level of acceleration excedence
doesn't do some quantitative damage, rather interesting experiment.
But these are all, in my view, things where you're looking
at do I have undue reliance on a single thing, whether that single thing
is reactor vessel rupture or, as happened in TMI-2, a cognitive error by
the operators that bypassed the whole event tree.
MR. GARRICK: One of the things that is kind of important in
that point about having undue reliance on a single thing is that there's
never a single thing even when it appears to be single.
By that, I mean, if you're talking about a reactor vessel,
for example, you have lots of things that give you indications of the
condition of that reactor vessel in terms of monitoring, etcetera.
So, it seems that, in those cases -- and the fuel cask
transportation is another example -- you may not have multiple barriers
in the classical sense, but in most of those cases, you have a great
deal more information about the -- its behavior.
If a cask -- we have seen it in tests at Sandia under the
most severe circumstances you can possibly imagine, and absolutely
everything was destroyed but the cask.
So, I think that, sometimes, that may be an
oversimplification, just because from a phenomena standpoint or from a
process standpoint, it may have that pinch point, and we have to offset
the vulnerability of that pinch point by additional levels of protection
that come in the form of information-gathering, diagnosis, monitors,
MR. APOSTOLAKIS: And all that means less uncertainty,
MR. GARRICK: Yes.
MR. BERNERO: Yes.
One could reformulate the whole system to say, rather than
undue reliance on a single barrier, you could have inadequate response
to a single challenge.
You know, you could restructure the whole thing logically to
MR. APOSTOLAKIS: We're interrupting you too much, Bob, but
counting the number of barriers has the same problem that in some
earlier times people were ranking minimal cut-sets according to the
number of events.
Ultimately, it has to come to the probabilities.
MR. BERNERO: Yes. And in reactor safety, I don't believe
you get there -- you have a regulatory system that gives you multiple
barriers rather prescriptively -- that is, reactor coolant pressure
boundary requirements, containment requirements.
It just doesn't give you the performance, and to resurrect
an old argument, you know, the regulations prescribe containment
performance predominantly as condensers for LOCAs rather than
respondents to loss-of-coolant accidents and core melts.
But anyway, one point I'd like to make on reactors is, when
you have a defense against some challenge, you need to have graded
You know, everything doesn't come out to the old PWR-1
release off-site, and I remember, years ago, in reactor licensing, we
used to have spent fuel handling accidents analyzed, and we consciously
used one-tenth of the Part 100 release guideline for analyzing a spent
fuel handling accident in the pool, which is almost a trivial analysis,
because you're under 20 feet of water and virtually nothing happens
off-site, and you have to look at that.
What are the consequences of the event?
When you get into material and waste, that becomes extremely
In material regulation, the concept of accidental release is
certainly with you, but chronic release and even deliberate release has
to be considered.
Exempt products -- I list there -- if you're not familiar
with the terminology in material licensing, when you go home and look in
the ceiling of quite a few rooms in your house, you'll see a smoke
detector, and the agency had a major deliberating problem in regulation,
because a typical battery-powered smoke detector has one-half of a
micro-curie of a 500-year half-life alpha emitter, americium-241, stuck
in there to ionize the air so that the smoke can cause an electrical
phenomenon that will make the little buzzer go off or siren or whatever,
the horn, and in regulating such a thing, you have to recognize, you're
never going to get them back.
They're not going to end up in a low-level waste or
high-level waste repository.
They're going to be thrown in the garbage. They're going to
be picked open by people. And so, you have to look at what I would call
chronic release and uncontrolled, routine release for things like that.
In order to have graded goals, you have to think through
what are the potential consequences of the act which you would
authorize, or the procedure, the barriers, protective actions, if they
are possible, and evaluate, a balanced choice of defense.
You can't prescribe it. It's far too complex. But as you
know, a lot of experience -- and you can bound consequences practically.
There are knotty problems. That's really a jurisdictional
In 1975, when the agency became NRC, there was the Food and
Drug Act that transferred patient safety for nuclear medicine to the
Food and Drug Administration, and ever since then, the states have
authority over patient safety, which is clear, but the NRC does not, and
It's really aside from here, although we had a lethal
accident about 1991. In Indiana, Pennsylvania, a brachytherapy patient
was killed by radiation, and the NRC requirements which were imposed on
that brachytherapy treatment had a device which reeled out wire with, at
that time, a four-curie source on the end of it into the patient's body,
and that device said I am now safe because I reeled the wire up.
The NRC required on the wall an alarming radiation dosimeter
and a personnel requirement that you would use a hand-held radiation
dosimeter in supplement. That was the defense-in-depth.
The source broke off. The machine said I got the source
back in its shield.
The alarming dosimeter went off, or it had gone off, and
stayed on. It was judged to be a false alarm, and they didn't use the
hand-held, and the lady died a very horrible death.
In that practice, there is a serious question, what is due
reliance or undue reliance on any barrier? What is the defense-in-depth
appropriate to that?
MR. BERNERO: Now, in waste, it definitely applies to
release barriers. As I said earlier, interjecting, the Nuclear Waste
Policy Act requires multiple barriers. So somewhere in a licensing
finding, somewhere in the licensing exposition by DOE, they have to show
the statutory requirement is satisfied because we have multiple barriers
and this is our demonstration of the adequacy of those multiple
barriers, as well as our performance assessment.
I underline the word "one" because the fundamental basis of
acceptability is not simply the total system performance assessment.
That's only one basis. You don't license to the safety goal.
There are other considerations that must be taken into
account. Some of these uncertainties are readily quantified, many are
not readily quantified. So you have to look at the whole body of
information in order to do it.
There is often confusion because defense-in-depth or
multiple barrier analysis is just another form of uncertainty analysis
and in this particular case, the staff, in Part 63 and in their
intentions for their review plan, have talked about guidance on how one
might do -- what's a sensitivity analysis, really, in supplement to the
appropriate uncertainty analysis in the total system performance
assessment, and I think that's good.
The one thing, and I talked to the ACNW in November, the one
thing that I think still needs attention is graded goals for graded
uncertainties. See, in high level waste, you deliberately put it out.
It's out there and now you're talking about what uncertainties do I have
about the barriers that inhibit the release and exposure of the public.
And one of the difficulties that exists is everyone that
talks about it seems to say the performance standard for exposure of
someone so far in the future, 10,000 years, 30,000 years in the future,
is such that it would not be greater than we would accept today, and
they come out and they use licensing acceptance criteria, which are
clearly acceptable. They're very low, they're very conservative.
There is no gradation of objectives to say, okay, well, how
far from the edge of the cliff am I, and I suggest that one can put
grades on radiation exposures from waste releases; that you can have the
clearly acceptable level of exposure, an acceptable level of exposure,
clearly tolerable levels of exposure, tolerable level on counting orders
of magnitude, life- threatening, and then clearly unacceptable.
And I have included a chart that I used before in November
and I just penned in. This is counting -- this is chronic doses and
then when you get to the top of the scale, you're really talking about
accident doses. For instance, when you get up to 10 rem, the accident
dose that's acceptable and has been for years, in things like reactor
accidents, 25 rem whole body exposure, is really a clinically detectable
What you're really saying is if you limit the accident dose
to 25 rem, that is a sufficiently harmless level because there are no
clinically detectable effects in the human body from that kind of an
exposure. You have to go up a factor of three or something like that.
I usually use 10 rem as that.
But when you get up in this high level we were discussing
earlier, you get up in cancer therapy, and you get doses like that. My
wife has just had very substantial doses.
So the whole point I'm trying to make, the focus is down
here. When you do the uncertainty analysis, it is nice if you meet your
clearly acceptable goal with your base case, but if you are depending on
some shaky uncertainty analyses, you should be looking for the edge of
the cliff; not only in uncertainty variation, but in objective or goal
variation, because you've got these orders of magnitude of tolerance
So that completes what I would like to say.
MR. KRESS: Thank you very much. Any questions, before we
move on the agenda? Very good. We are now at a point in the agenda
that calls for a general discussion of the people at the table and
anyone in the audience who wants to join in, and we need to define the
issues for further consideration.
I don't know exactly how to approach this, except ask for
any volunteers that want to make additional points or question the
MR. APOSTOLAKIS: If I could make a suggestion. Why don't
we start out by defining perhaps three or four or five points that need
some discussion, because otherwise we will be going in ten different
MR. KRESS: That's a good suggestion, George. Do you want
to make a stab and give us a couple of points?
MR. APOSTOLAKIS: Well, this issue of uncertainty that I
raised, I think, deserves some discussion and whether we want to place
defense-in-depth in that context. That's certainly something that I'm
MR. KRESS: That's a good one. What I'm interested in, of
course, is the issue of should there be a specified allocation.
MR. APOSTOLAKIS: That's a good point.
MR. KRESS: That would be one.
MR. APOSTOLAKIS: And I must say I am still not comfortable
with my understanding of the issue of how to use defense-in-depth in the
high level waste repository. So maybe a summary of the issue and then a
discussion, a summary perhaps by John, would help me understand.
MR. GARRICK: One of the points I'd like to see on here,
too, we keep hearing this observation that licensing decisions should
not be based on PRA/TSPAs alone. I'd like to see us discuss that more.
MR. APOSTOLAKIS: Okay. That's a good point.
MR. KRESS: Yes, that is, particularly when we're talking
about entering into a mis-conformed regulatory system. That's four
pretty good items. Are there others people would like to add to the
list? I think those are a pretty good set of things.
I would like to add one more, and that is we have heard some
contrary and different opinions on this. Should we have -- well, we've
been calling them safety goals, but I've been calling them risk
acceptance criteria that we regulate to.
Should we have risk acceptance criteria that we regulate to?
MR. GARRICK: And I don't think, by the list here, that we
would want to bound up anybody from jumping the fence here.
MR. KRESS: Absolutely.
MR. GARRICK: If they have a burning issue that they think
is critical to the subject.
MR. KRESS: Okay. That's, I think, five pretty good issues.
How should we approach the discussion of these? George, do you have an
idea on that? Would you like to, say, take one and I take another one
and John take another one and --
MR. APOSTOLAKIS: Sure.
MR. KRESS: -- just throw out some thoughts and see what
kind of response we get?
MR. APOSTOLAKIS: We could do that, yes.
MR. KRESS: Why don't you start with the issue of
MR. APOSTOLAKIS: Okay. Well, I tried to make a case
earlier today that the reason why we are revisiting the issue of
defense-in-depth is that we can now quantify a good part of the
uncertainties associated with the performance of the systems that we're
talking about that we could not quantify 15, 20, 30 years ago.
That includes identification, quantification,
characterization, all the words.
I also made the point that the language is extremely
important here. I was glad to hear Tom Murley say that, in his mind,
defense-in-depth has always been a philosophy and not a principle,
although the word principle is being kicked around. But I think Bob
Budnitz's point is well taken, that it ultimately comes down to what you
I mean, what you call it is nice to have good terminology,
but what you actually do at the lower level, at the working level, is
what counts, and that's what I want to address.
I really think that for the uncertainties we have
quantified, defense-in-depth, the words don't belong there. You're
going to use the tools of defense-in-depth, barriers, diversity and so
on to manage your uncertainty and you have an excellent means, a
numerical standard against which you can decide how much is enough,
which is really a fundamental question today, how much defense-in-depth
MR. KRESS: But, George, we don't have numerical standards
on how much is enough, unless you allocate --
MR. APOSTOLAKIS: Yes.
MR. KRESS: Now, if you would throw in this word allocate, I
would agree with you. But then, by my definition, that becomes
defense-in-depth in a regulatory sense, if you allocate.
MR. APOSTOLAKIS: But I would avoid the words defense-in-
depth, because they carry a certain baggage. Now, I understand where
you're coming from and in an ideal world, but I want to reserve the
words defense-in-depth to mean what they have meant all along; handling
unquantified uncertainty by using barriers, emergency plans.
MR. KRESS: Let me give you my problem with that. I
mentioned I my talk that I don't think we can live with unquantified
uncertainties in a defense-in-depth regulatory system. The reason I
said that is I don't know what to do, I don't know how to put limits on
defense-in-depth, I don't know how many barriers I need, I don't know
how good they have to be, I don't know where to put them.
And then when I do this, I don't know how well I have
compensated for the unknown uncertainties, and I'm saying you really do
have to have some knowledge of what that level of uncertainty is and how
putting barriers in different positions will compensate for it; how much
of that uncertainty will you get rid of or will you lower your achieved
risk to a level that that uncertainty is acceptable.
So I'm saying you really do need a quantification metric in
this, even for what we're calling unquantified uncertainty.
MR. APOSTOLAKIS: Okay. My response to that is, first of
all, the problems that you delete and the problems that you just gave
us, I would say that's the price you pay for not quantifying
The second is, again, one of my bullets said that if we do
that, we will focus attention on unquantified uncertainty, and then my
hope is that by doing that, we will eventually do what you're saying,
because somebody might say, well, gee, is it really unquantified. Maybe
we can have an estimate of the probability that all this is wrong, but
right now we don't do that.
Therefore, right now, you pay the price. You put the
barriers and you pay the price. I'm sorry, what?
MR. BERNERO: I'd like to interject on this. In the earlier
discussion, we talked about if you quantify the uncertainties, you could
make a case to eliminate the containment, say, on a class of reactor.
MR. APOSTOLAKIS: Right.
MR. BERNERO: Setting that aside, if, on the other hand, and
to Tom's point that I've got to know what to require, like some
prescription, consider, for the moment, if one would resurrect the
question of urban siting of reactors, because of the growth in the
United States and the availability of industrial property, getting close
to load centers, now, that is almost impossible to quantify the
uncertainty associated with that siting ramp.
And it's an interesting thought experiment to say what
quantification of uncertainties or what formulation would be appropriate
to reconsider that. I don't think you can do it by having a regulatory
agency invent a new siting policy, saying here exactly are the
population distribution criteria and everything that we would have to
set rational bounds on it.
If you go back to the 1980s, the late `70s and early `80s,
the agency was very heavily involved in a siting study or a series of
siting studies to attempt that.
MR. KRESS: I'm going to make a provocative, radical
statement, so everybody knows that that's what this is when I say it.
I basically think the Europeans have the right idea that
it's irrational to rely any at all on emergency response to meet risk
acceptance criteria. Now, that's a radical, provocative statement, but
I think it is irrational. I think it's part of the whole problem of why
there is lack of public acceptance in nuclear power.
And if you could design into the system to meet risk
acceptance criteria at an acceptable uncertainty level, without
requiring emergency response, then I think then emergency response
becomes a true defense-in-depth, because you're not relying on it to
meet your risk acceptance criteria. You're just saying suppose we're
wrong, let's have it anyway.
MR. BERNERO: But you aren't now.
MR. KRESS: I know. You don't meet risk acceptance criteria
without emergency response in this country.
MR. BERNERO: I don't agree with you. Reactor siting
studies that were done in the late `70s and early `80s, it is there as
defense-in-depth, but you didn't have to meet it on emergency response.
MR. KRESS: I do not think you will meet the safety goals
without effective emergency response. This is a point we'll agree to
MR. BUDNITZ: I have a puzzle for you, staff and ACRS, that
I can put in a pretty stark context. I want you to imagine you're
running a reactor in one of the former Soviet countries. Soviet's gone,
but there were, of course, several countries, Lithuania, Armenia,
Russia, Ukraine, that are running reactors, and a lot of those don't
have a containment at all. The old 442- 30s certainly are BMKs.
The United States Government, as a matter of policy,
implemented through the Department of Energy and the State Department,
has, as a policy, that we are trying to get those governments to shut
down all of those reactors as a matter of our policy. We have stated
that to them at the highest levels and it's part of our detailed policy,
too, I know, because I work in this arena a lot.
So that, for example, Richardson is going to go to Lithuania
in February. He is likely to tell them that we continue to oppose
running Ignolena and RBMK because it's not safe enough.
Now, suppose a government there says we've done a PRA.
Suppose a water reactor, not an RBMK, where the PRAs are more reliable,
and the core damage frequency is several times ten-to- the-minus-four,
but considering our desperate economic situation, we need that reactor
and that's safe enough for us.
The U.S. Government policy position today is no containment,
shut them down. By the way, it's not the only reason, but no matter
what else you do, no containment, let's say for the 442- 30s, whatever,
What do you think of that? Knowing as much as we, everybody
around this table that knows reactors knows about them, about what those
probabilities mean, knows what -- and you understand the government says
we're going to take a bigger risk than you would be willing to take in
the United States because we need the power, that's their prerogative,
as a matter of sovereignty, and they say we know it's not contained, we
know that the consequences were we to have one of these would be greater
than they would be in the United States for a water reactor of the same
They have said that one crucial element that we invoke of
our defense-in-depth philosophy, as implemented through the containment,
is absent and is still acceptable.
Now, I'm not arguing about their right to make that, that
they're sovereign, but what about that here, what would you say?
MR. APOSTOLAKIS: It's a different objective.
MR. BUDNITZ: I understand that, but what do you think --
MR. APOSTOLAKIS: So it's not an issue of defense-in-depth.
MR. BUDNITZ: But what do you think about whether -- suppose
they were three-times-ten-to-the-minus-seven and 440 megawatts, would
that be acceptable in the United States without a containment? No, not
today in the regulations. But what do you think about that as a matter
of whether it should be?
MR. APOSTOLAKIS: There's nothing we can do about it.
MR. BUDNITZ: No, no. But in other words, we're at three-
times-ten-to-the-minus-seven core damage frequency in the United States,
440 megawatts, would that be acceptable here to you?
MR. KRESS: The question would it be acceptable or not is a
tough question to ask, because it's a judgment to be made on --
MR. APOSTOLAKIS: It's a policy issue.
MR. KRESS: The question is whether it's a rational position
to take, a different question, and I think it's entirely rational to say
that that's a reasonable position to take. As long as you state your
goals on what risk acceptance criteria you're willing to live with in
terms of the uncertainty and its determination.
If you meet that ten-to-the-minus-whatever at a level of
uncertainty that's acceptable, then it's a perfectly rational position,
and that would be the rationalist view of defense-in- depth.
MR. BUDNITZ: I heard you expound that, and George saying.
On the other hand, I heard my close friend Tom Murley say, and I think
I'm with you here --
MR. APOSTOLAKIS: Unlike me, you mean?
MR. BUDNITZ: No, no. You're another close friend. But Tom
said, and he's sitting here, so maybe he -- he's two meters to my left,
so he'll say what it he wants for himself; that no, no, in the United
States, we wouldn't like a reactor without a containment, just totally
MR. KRESS: That's another question. I think it's probably
true, we wouldn't like it.
MR. BUDNITZ: I'm not saying whether we wouldn't, not
whether we wouldn't, but whether we should.
MR. GARRICK: I think it's a bit irrelevant. I think it is
a policy question. First off, at these reactors you're talking about,
if I had to make that judgment, I would -- getting back to George's
topic -- I would really want to turn up the microscope on the
uncertainty of the core damage frequency.
MR. BUDNITZ: Of course. I wasn't arguing that case.
MR. GARRICK: And I think I would find the kind of
information that would suggest to me that the U.S. policy is sound.
MR. BUDNITZ: I'm not arguing that for a minute. I
subscribe to that policy.
MR. MURLEY: John, could I make a point, too?
MR. BUDNITZ: Of course.
MR. MURLEY: Coming from the outside now, there's almost an
air of unreality to this discussion, because you've got to take into
account the human safety culture issues, which do cut across a lot of
these sequences and stuff.
MR. BUDNITZ: Of course.
MR. MURLEY: So Bob's premise, I think, is unrealistic. I
agree if you could absolutely prove that you had five times or
four-times-ten-to-the-minus-seventh or something, but I don't think
anybody believes you can ever do that with humans.
So you just have to keep that in your discussion somehow. I
think I understand what you're saying and the premises and so forth, but
the public, listening to this, think that what were these guys -- what
do they own, what do they have.
MR. GARRICK: I would like to comment to the allocation
issue, because I think it's --
MR. APOSTOLAKIS: That's another issue.
MR. GARRICK: Well, we've drifted into it from talking about
uncertainty. I've got plenty to say about that, too.
I need to understand a lot better, Tom, what your bounds and
references are with respect to the issue of allocation. But on the
surface, it bothers me a great deal.
The reason it bothers me is that the risk assessment is, in
my view of a risk assessment, a set of scenarios and the performance of
a particular system that you may want to allocate some risk criteria to
is strongly dependent upon where that piece of equipment sets in what
I'm sort of reminded about the situation following the Three
Mile Island accident, when there was all this fuss about maybe we should
add a third auxiliary feed water pump to all of the reactors.
So there was an analysis that was performed as to what
benefits you would get from adding that third auxiliary feed water pump.
The answer to the analysis was that, well, if you added, in the context
of what the NRC views as a safety grade auxiliary feed water, the
benefit is very marginal. But if you remove the NRC criteria and are
allowed to not have that auxiliary feed water system have to depend on a
coolant system, a chilled water system, get it out of a hard room, so to
speak, and put it in something like the turbine building, where you
don't have to rely on certain support systems, you get a heck of a lot
And I can point to hundreds of those kinds of examples in a
nuclear plant, and so I have a great deal of difficulty knowing how you
could possibly allocate risk criteria in a situation where you have
reactors and plants as different as they are, where you have accidents
extremely dependent upon -- or the performance of systems extremely
dependent upon where they fit in the accident sequence.
And that may not be what you're talking about, but it's
something that bothers me. And I think that one of the things that's
fundamental and crosses a lot of these issues is that we're still
learning and the safety goal issue only began to formulate some meaning
after we started to get some results of risk assessments.
I remember the Commissioners arguing about -- and it was a
ridiculous argument -- about whether it should be one-times-ten-
to-the-minus-four or five-times-ten-to-the-minus-four, on a parameter
where the uncertainty is a factor of ten.
That's why the uncertainty is so absolutely critically
important here. As one of my colleagues would say, the uncertainty is
the risk. That's where the ballgame should be played.
I've never been one to think in terms of uncertainty being
complimentary to risk, but rather uncertainty being an inherent element
of risk assessment, just as I would argue, and that brings me down to
the TSPA/PRA issue and how much we should depend on it, that if we can
think of something in addition to the TSPA or the PRA that's a basis for
decision-making on the safety of the plant, we damn well ought to be
bringing that into our risk assessment and our TSPA.
Expert opinion, for example, is not something that should be
outside the scope of a risk assessment. So we should be striving in
that regard to make the TSPA and the PRAs as encompassing as possible.
Now, when the NRC got into the PRA act and was trying to
respond to the criticisms of the industry that they were too expensive
and went to a highly simplified and limited scope, and as the image
started to develop, in people's minds, that a PRA was something much
less than what it might be, then I can understand why you would have to
conclude that you've got to consider things beyond what's in a PRA, if
by what's in a PRA is what the NRC meant by the old IPE, where there was
essentially no uncertainty, no external events, and not much scope.
So I think these are things that really make it very
difficult for me to imagine how we can get unduly specific with respect
to something like allocation.
MR. KRESS: Let me respond a little bit to that. You can
envision all sorts of levels of allocation. You could allocate system
reliability or even component reliability. That's not what I had in
mind. I think basically with defense-in-depth, we're dealing with
prevention versus mitigation. That's basically what we're doing.
The four elements of that I talked about. What I had in
mind here was let's take the case of nuclear reactors, power reactors.
We're talking about core damage frequency versus conditional containment
How are we going to allocate between those two to meet, say,
LERF, which is our overall thing. What I'm saying is that in decision
theory, you ask the question if a core damage manifests itself, what are
the consequences of that in terms of my loss function; how valuable is
it to me to prevent that from happening, as a regulatory agency.
You've got to make a decision theory process and you arrive
at a loss function that says that's so valuable to me that I want to
place goals on core damage frequency or risk acceptance criteria, and
there are probably going to be a lot more going into the prevention than
there is to the mitigation.
Then you also ask yourself, well, suppose you do the same
thing with the conditional core damage frequency. You take another loss
function. What is -- and it basically becomes what's remaining of LERF,
because you've already established the loss function with your CDF.
That's a level at which I would advocate the allocation.
MR. GARRICK: Well, that's what I said, I qualified my
comments with not knowing what you really meant by criteria.
MR. APOSTOLAKIS: But in this context, then, when you talk
about, first of all, prevention and mitigation, in this case, are terms
with respect to core damage.
MR. KRESS: Yes, absolutely.
MR. APOSTOLAKIS: Because you are preventing the release of
radioactivity to the environment. In this sense, then, there is no
prevention in performance assessments. It's all mitigation, isn't it?
It would be released from -- no? What are you preventing?
MR. BUDNITZ: If you can keep it inside the canisters, long
as it's inside the canisters --
MR. APOSTOLAKIS: For 10,000 years?
MR. BUDNITZ: If you can keep it inside the canister for
10,000 years, that's prevention. I would -- in other words, it hasn't
gone anywhere. That is, in fact, the case for canisters that we talked
MR. GARRICK: If you can keep the water away, you can show
MR. BUDNITZ: So, George, I see that break between
prevention and mitigation as very hazy for Yucca Mountain, but I
certainly know what prevention means. Prevention is keeping it from
going anywhere. It's just in the can.
MR. BERNERO: I beg to differ on prevention. The inherent
act of waste disposal is to place the material in the biosphere or
geosphere and from then on, the performance assessment is modeling what
MR. BUDNITZ: Right.
MR. BERNERO: Does it stay in place or does it ever so
slowly corrode, decay or whatever, and there are features in waste
disposal systems that can enhance, say, containment performance.
If Yucca Mountain adopted, as I wish they would, the
addition of depleted uranium filler in the container, I think that would
greatly enhance --
MR. KRESS: That would be a wonderful addition, I agree with
MR. BERNERO: Yes. But, see, this is the thing. You're not
preventing something, you're inhibiting it.
MR. BUDNITZ: That's fair.
MR. BERNERO: And I think there's a danger -- it's really a
barrier, an inhibition to the movement of the waste, because that is the
measure of performance.
MR. BUDNITZ: Yes, but when we talk about prevention in a
reactor, we mean keeping it inside where it started. In that sense,
it's not a perfect analogy, but it's not such a bad one to say that
prevention is -- the earliest state -- keep it inside the can.
MR. KRESS: I also added -- in my definition of prevention,
I added the word intervention and you have lots of time and lots of
intervention strategies one could choose. So I would say there is --
MR. BUDNITZ: Except as a matter of public policy, the NRC
has said that they're not going to count on any human intervention 6,000
MR. KRESS: I know, but that's a policy statement.
MR. BUDNITZ: I understand that.
MR. GARRICK: I think I can make one observation that covers
a lot of my concern here about issues of allocation and definitions and
what have you, and it has to do with I don't think we should do anything
that bounds our thinking about the safety of what we're dealing with, be
it a repository or a reactor plant.
We all know that we've had experience with this. When we
adopted the design basis philosophy of safety of nuclear power plants,
we, in a sense, bounded our thinking. The game became if you come
forward with a design basis accident and you convince everybody that
it's acceptable, then you're okay. It's the same thing. The other
language we've heard about is beyond Class 9 accidents.
There shouldn't be those kind of artificial thresholds and
boundaries, even though it made it more convenient, from a regulatory
standpoint. And allocations have a tendency to do that and subsystem
requirements have a tendency to do that. They have a tendency to narrow
the view of what we should be analyzing, what we should be designing
against, and what we should be analyzing, what we should be designing
against, and what we should be controlling.
Even core damage frequency is a limitation, because I can
think of scenarios in lots of plants that would decrease the core damage
frequency and increase the public risk, and I think we have to be very
open and clear about that, and I think that's the virtue of PRA.
MR. APOSTOLAKIS: I disagree, though. I think there is an
element that's missing here.
MR. GARRICK: You disagree?
MR. APOSTOLAKIS: No. It's not -- when we say allocation,
we should not take it only in the mathematical sense that you want to
have a certain -- meet certain goals and that you allocate the
performance of various systems. There is a more fundamental reason why
the staff wants to do some of that.
Even though there may be situations where you are -- you
know, a certain measure, as you just said, may decrease or increase the
core damage frequency, but the role is beneficial, the staff wouldn't go
for it, because core damage by itself is an undesirable event.
See, the assumption in what you said was that all I care
about is the QHO and the staff will tell you no, that's not all I care
about. In fact, the new oversight process makes it very clear in black
and white. The staff says we care about initiating events, we don't
want to see any of those. Why? Well, they aren't going to put it on
paper. They will tell you, though, that they don't want to be on the
front page of the newspapers. We don't want to see the primary system
Why? It creates public outcry. We don't want that. So
there are more objectives that perhaps have not been spelled out in the
books until recently for which -- which you are trying to meet, and if
you look at it that way, then you are saying, well, maybe core damage
frequency is something I worry about, because it's not just a QHO.
The fundamental question is, though, whether you have
similar situations in the performance assessments and I think one of the
reasons why you don't is time.
In reactors, we can have a problem tomorrow with an
initiating event. In your case, you're talking about thousands of
MR. GARRICK: Yes, the conditions are entirely different.
The real issue of risk probably in the waste field is the operational
risk and the handling and the way in which you do things.
MR. APOSTOLAKIS: But my point, John, is that maybe the word
allocation for reactors is not the right word, because they are not
allocating anything. They are saying I don't want this to happen, I
don't want the core damage event, I don't want an initiating event.
MR. KRESS: When I say allocation, I mean I don't want that
to happen at this frequency, with this uncertainty, with this confidence
MR. APOSTOLAKIS: I understand that.
MR. KRESS: That's what I mean by allocation.
MR. APOSTOLAKIS: But there is a reason why they don't want
it to happen, because that by itself is bad; not only as a contributor
to core damage, but if I have a LOCA tomorrow, the agency doesn't look
MR. GARRICK: But, George, you're not saying that the NRC
disallows the core damage. They can't do that. They can't do that.
Are you saying that -- what you seem to be suggesting is that the NRC
really doesn't think in terms of a ten-to-the-minus- four core damage
frequency, but a ten-to-the-minus-infinity.
MR. APOSTOLAKIS: When did I say that?
MR. GARRICK: Well, you made the point that they wouldn't
accept it. Well, what are they not accepting? They can't stop it.
They can't stop the fact that the core damage frequency has a likelihood
MR. APOSTOLAKIS: What I'm saying is when we say allocation,
we have to be very clear what we mean. That comes back to what my
objectives are when I regulate. I got the sense from your earlier
comments that what you thought was the objective of the regulation for
Yucca Mountain or for reactors was the ultimate quantitative health
objectives or, in Yucca Mountain, the dose. The ultimate criteria, in
And then allocation, in that sense, means that some engineer
says, well, gee, you know, this is really my objective, but I would like
to see this performance here, that performance there, in the system.
What I'm saying is, no, there is a fundamentally different view of
regulation for reactors. It's not only the public health and safety.
That's how we start, but that's not our only objective. We
don't want to see core damage events by themselves, even though they
don't affect public health and safety, because they're contained.
But even more than that, in fact, the staff said it very
clearly, the initiating events, we don't want to see too many of those.
They create those sorts of headaches, other things. We don't want to
see -- whatever -- the four cornerstones they have. So what I'm saying
is that the decision problem is different in this case in the sense that
I have different objectives and I'm not allocating anything anymore.
All I'm telling you is I really don't want to see this.
MR. LEVINSON: But, George, I think historically we have
confirmation. The importance of TMI was not exposure of the public.
The importance of TMI was that it was core melt.
MR. APOSTOLAKIS: Yes. Yes. And we saw the reaction and so
on. So that supports, in fact, the staff's position. You may have -- I
mean, as Tom said earlier, you can have a TMI every year and you still
meet the goals. You tell me who at the NRC would accept that.
MR. GARRICK: And my only point is be careful about the
blinders you put on to support the staff's position, because we put
blinders on us to support the staff's position in the past and we
probably should have not. Be careful about that.
MR. APOSTOLAKIS: I'm not sure they're blinders.
MR. GARRICK: Well, you're the one that's suggesting that.
I think that all I'm suggesting, all I'm suggesting is that the real
virtue of the risk thought process, and by which I mean all these things
we've been talking about, quantification of uncertainty, complete set of
scenarios, doing the best possible job we can, is that we have not built
ourselves artificial thresholds, like safety-related systems.
I think that that's the thing that is an important virtue of
it that we should not lose by adding some constraints.
MR. APOSTOLAKIS: And I agree that they should not be
artificial. But look what happened at Northeast Utilities. Was that
artificial, was that a real reaction? Was public health and safety
threatened at any time?
So it's clear to me that for reactors, it's not just public
health and safety.
MR. GARRICK: Well, I agree with you and I want to stop
because I want to hear from a lot of people. I would say one of the
greatest advances we've made in the improved performance of the nuclear
plants in this country is not the business of the traditional safety
analysis and what have you, but it is the emphasis that the utilities
have been giving to human performance.
I am really impressed with what you will find at most
utilities today on evaluating human performance and how to motivate them
and how to challenge them and how to make them accountable for what
they're doing. And it's true, in the sense that it's outside our
database, which it isn't totally, we don't consider a lot of those kind
MR. LEVINSON: If I can make just one more comment, John. I
think these are not at all inconsistent. The value of good analysis to
reduce uncertainty, PRAs, et cetera, certainly is something we should
all strive for, but I think the point is what we get from it is not just
a single number, like dose to some person in the population.
It can also be used to achieve other objectives, like
reduced core melt. So the fact that you might have multiple objectives
for the PRA is not inconsistent with depending on PRAs and proving them.
MR. BUDNITZ: Let's go to Yucca Mountain for a minute. When
Part 60 was under development, I was on the staff 20 years ago when we
were thinking hard about it, and at that time, nobody had confidence
that what we now call performance assessment could be good enough to be
relied on as a principal means for understanding. And because of that,
the staff, at the time, wrote the subsystem performance requirements,
the canister lifetime and some canister leakage rate per year and the
thousand year travel time and so on into the regulation.
Notwithstanding everything else you did, you had to show
this thousand year travel time, for example. The staff explicitly, in
the statement of considerations of Part 63, just this year, said 15-18
years have passed; we now, says the staff, and I agree with this
entirely fully, we now have the confidence in the analysis methods and
the data that we didn't have them, we the same staff or the different
folks of the same staff, and, therefore, we feel that those things have
been superseded by this new technology and its use and our confidence in
So they have come to the stage where they used to have what
you'd call barrier -- the concept of these multiple, whatever else you
do, you've got to do barriers or something, performance, they've
abandoned it for the moment.
I mean, there's still this other thing, and I think that's
completely correct. When evolution of knowledge enables you to say I
now don't have uncertain values to have, I now can do certain analyses
and I can have confidence in them at a certain level, I no longer need
what I used to need 18 years ago. That is completely rational.
MR. APOSTOLAKIS: But your objective is still to meet the
dose criteria. I fully agree with that approach. You don't have any
intermediate objectives. So what I'm saying is that in reactors, it's
MR. BUDNITZ: No, no. I'm not -- of course, I'm not arguing
with you for a minute, but then all of a sudden, in the same statement
of consideration, Part 63, they say but besides the dose objective in
Amergosa, we have this defense-in-depth. My slide showed, I asked the
question, well, if we're going to invoke it, can they flunk on
defense-in-depth, even if they meet that other thing with lots of
margin, and apparently the answer is yes.
The staff has said yes, they could flunk on defense-in-depth
and then you have to ask, well, what does that mean. I was trying to
probe in my slides what that might mean in terms of some sort of
allocation or in some sort of a figure or in some sort of a do it
analysis of a degraded or under-performing barrier and tell us what it
means and whatever it means, are we going to flunk you on that one.
If Yucca Mountain can flunk on one of these, even though
they meet the overall thing with lots of margin, then you have to figure
out what does it mean, what sort of allocation have you come up with,
MR. APOSTOLAKIS: You just said that now we have confidence
that we can calculate these.
MR. BUDNITZ: It's not a perfect tool.
MR. APOSTOLAKIS: But let me ask you this. What are the
major unquantified uncertainties in performance assessment?
MR. BUDNITZ: Unquantified uncertainties.
MR. APOSTOLAKIS: Yes.
MR. BUDNITZ: I suppose they'd be some of the models that we
still haven't tested well enough.
MR. APOSTOLAKIS: This is not something people talk about?
MR. BUDNITZ: Of course, we talk about it every day.
MR. APOSTOLAKIS: So models --
MR. BUDNITZ: It's at the center of what we talk about.
MR. APOSTOLAKIS: Are these uncertainties large enough to
invalidate the performance assessment itself?
MR. BUDNITZ: Well, my personal view is that Yucca Mountain
is very likely to meet that dose criterion out there in Amergosa with
lots of margin, including these.
MR. APOSTOLAKIS: Including the unquantified.
MR. BUDNITZ: Including -- I mean, there is some judgment
about the models. You always have to bring some judgment in the end,
because not everything has been tested, especially with those long
timeframes and that's certainly true of the metallurgy of the can.
But it is my view that in the end, that will be the case.
I'm still holding open judgment because the final design isn't here and
certainly analyses haven't been done on that. But if that's true, if it
turns out that there's lots of margin against the dose, the staff says
but you can still flunk because you flunk something about
defense-in-depth, what is that?
I'm struggling with it, because it isn't the same as what
you're saying, well, a core melt is bad. You know, Millstone was bad.
It's not the same sort of thing.
MR. APOSTOLAKIS: I understand that. That's what I keep
saying for the last ten minutes. They are two different things. If you
guys knew, if the Commission believed that by building Yucca Mountain,
you will have a major incident five years later, I'd bet you there is
going to be an objective there in order to have it.
MR. BUDNITZ: Yes, of course, or --
MR. APOSTOLAKIS: If it's a thousand years --
MR. BUDNITZ: Or even if it's a thousand years, because they
have a 10,000 year criteria. So I think it's a challenge. I'm looking
at Ray and John from the ACNW and all of us that have thought about this
hard. It's a big challenge to figure out what you mean and what you do.
MR. APOSTOLAKIS: I see there are two different variables.
MR. BERNERO: Tom, I'd like to interject here. The
discussion of an incident in the near term against a waste disposal and
also a remark that John made earlier about if you've got some
significant uncertainties, get them into the performance assessment,
which is an admirable objective.
First of all, there has to be not an allocation, in my mind,
but a recognition that in waste management, and I will use low level,
near surface waste disposal, as an example, there is a sequence of
allocated allowances or decisions; is this site acceptable, is this
emplacement design going to be an acceptable compliment with the site,
and, of course, taking the whole system into account, is it going to
satisfy the performance assessment requirements, the dose limits
off-site and so forth, taking account of the uncertainties and climate
and flow and intrusion and so forth.
Now, if you go, as a practical matter, in Part 61, there are
explicit site criteria and there is an extensive body of guidance on
performance assessment, but there is not a good way to analyze, to do
the uncertainty analysis of emplacement techniques.
Basically, what any new site that was going to be built east
of the Rocky Mountains, what they did is just adopt the French approach,
and the French approach is select the site that's proper, build it with
dual liner leachate collection system caps and all the bells and
whistles, and do your level best to make sure it never leaks.
And you don't quantify that in the performance assessment.
You have uncertainties and you live with those uncertainties. Take the
item 129, if you go to a low level waste disposal site, all these
shipments that come in, and you're talking 100,000 shipments, big
numbers, they all have item 129 is less than or equal to X.
It's detectability limit and if you take 100,000 times less
than or equal to X, it's five orders of magnitude higher than that.
I've had the authority for the French low level waste site at Loeb tell
me that halfway through, we're going to hit the limit on item 129, and
he doesn't have a performance assessment technique to get out of that.
He doesn't have an analytical detection technique. He's got to use some
And ultimately I think they will get out of it. They're not
going to stop and say this is the limit for this site, because it's not
real and it's also not a real threat, item 129.
So there are many things in waste disposal that you cannot
firmly quantify. You've got to evaluate and make a judgment. It's very
difficult. And the decisions, right now the staff is heavily involved,
and the Commission, too, in advising or concurring in what DOE is doing
to clean up its waste tanks and a high level waste tank, when you
extract that waste, the Commission promulgated criteria on how can you
stand up and say the high level waste is out, when you know there is
The residue isn't well quantified, it isn't well located,
and it's the difference between two very large numbers and it's very
difficult to do uncertainty analysis on it.
You can't characterize it, you can't sample it. And so your
performance assessment for that site is going to say I'm satisfied that
you've extracted enough, DOE, and that you have made a persuasive case
about how you grouted it, how much grout there was, how much residue you
estimated it to be, and so forth, and then you're going to do a very
elementary or simple performance assessment that doesn't take any real
credit for the grout and the can and many of the barriers.
MR. KRESS: This is an interesting discussion, Bob, because
I think what you're saying is here is a circumstance where we just have
uncertainties that we can't quantify, so what we do we do in that case,
in a risk-informed regulatory world.
MR. BUDNITZ: That will be true at Yucca Mountain. There
will be some uncertainties we can't quantify.
MR. KRESS: So it's an interesting question, what do you do
when you can't quantify the uncertainties. I think you fall back on
arbitrary defense-in-depth. Arbitrary in the sense that you put the
best you can here and there.
MR. GARRICK: You fall back on a combination of some sort of
MR. KRESS: I want to just introduce a conceptual note here,
because what you're really saying, Tom, is that it's not so much you
can't quantify it, but you just don't like the result, because the
principals ought to be there, that you can always quantify it. It just
may be that you have ten orders of magnitude of uncertainty when you
would like to have two.
And in the presence of that level of uncertainty, then you
have to do something. But I think that the whole discipline that we're
talking about here is to be able to assign values to parameters based on
the evidence that you have, and you always have some, but in the
problems we're dealing with, there are too many areas where we have much
less than we'd like.
One of the things I would like to do here before the break
is look to my colleague, Ray Wymer, on the performance assessment angle,
who has been doing a lot of thinking lately about some of the key
uncertainties associated with one aspect of performance assessment
that's critical to improving the models, and I suspect, Ray, you could
identify some examples of areas of uncertainty on the chemical side and
offer opinion about the likelihood and what needs to be done to resolve
Would you comment on those and kind of against the
MR. WYMER: I suspect you think I've been too quiet for too
MR. KRESS: Yes. I know you have a lot to say and I hope
that there is an opportunity for you to do so.
MR. WYMER: I'll say a little bit about chemical
uncertainties, which is fairly specific, and then I think tomorrow, when
we adjourn discussion, I want to make some general comments that I've
noted down here that are not necessarily appropriate to this specific
discussion we're having right now.
But there are a lot of chemical uncertainties with respect
to Yucca Mountain and the repository. For example, there still is
uncertainty about the corrosion behavior of alloy C-22 and while there
is a lot being done, it still remains that you can't take a couple of
years worth of studies and extrapolate them for 10,000 years very well,
although the more basic understanding you have, the better off you are
in your extrapolations.
So the primary line of defense, which somebody mentioned,
maybe Bob Budnitz, that the waste package, the waste container is really
the principal reliance, which is true, for containing the waste and
preventing exposure, there is uncertainty remaining there, which people
are working trying to narrow, both in the NRC and in the Department of
In addition, there's a good deal of uncertainty about the --
once you breach containment and you get into the fuel material itself,
there is a lot of uncertainty with respect to the formation of secondary
precipitates, materials that would tend to provide another line of
defense against release of radioactivity.
People don't really know what these second phases are. They
are extraordinarily complex because of the complexity of the nature of
the fuel and the nature of the corrosion products that meet that fuel
and the complexity of the water that's coming in.
So there may be additional barriers to release. There's a
lot of uncertainty there, though, and there's been no real attempt, no
real concerted attempt to quantify those processes that may limit
release of radioactivity in a significant way.
It's been mentioned briefly here that you can put in
backfill materials, like UO2, into drift, or you can actually put those
inside the waste package, which, by a saturation effect, can reduce the
rate and extent of dissolution of the fuel, and also lead to additional
secondary phase formation.
These are all uncertainties. Most of what I mentioned, with
the exception of corrosion, is an uncertainty at the direction of
greater containment of the radioactivity to make the waste environment
more retentive than the analyses are currently showing.
But without belaboring the point too much, there are
chemical uncertainties which are, in my view, large. There are a number
of mitigating things that could be explored, like backfill materials,
that could enhance the safety of the repository and could decrease
somewhat the uncertainty in the analysis, and all of these things, in
the best of all possible words, would be examined.
The time constraints that we have with respect to the
license application would seem to pretty severely limit the amount of
investigation you could make of some of these potentially very important
chemical thought processes. However, if, for some reason, we get into
the bring-me-another-rock mode, there may be more time available to
solve some of these problems.
MR. APOSTOLAKIS: Are these uncertainties in the PA's now?
MR. WYMER: Only in a very general way, George. There is
practically nothing that I could think of or that anybody could think of
that hasn't been mentioned in the performance assessment, but mentioning
them is one thing and dealing with them competently and comprehensively
is quite another thing, and I think it's that latter that's weak.
MR. GARRICK: One of the things that's very interesting
about these problems, I'm always looking for comparisons. The key to
the reactor safety problem is water. The key to the safety problem is
the absence of water. Also, it turns out that one of the attractions of
using core damage frequency as a measure of performance in the reactor
is because of the step change in uncertainties that occur once the melt
occurs, and you try to quantify the accident progression.
But we're kind of in that position in the waste field. We
have a problem that's not too dissimilar in terms of the bounding of the
problem and what have you. Fortunately, the time constants are much
longer and that's to our advantage, but the problem in the waste field
is once you get the material mobilized, coming up with models that do a
rational, reasonable job of defining the mobilization, the retardation,
the dilution and the transport of the radioactive material.
It's a problem not too unlike the accident progression
following core melt, although the thermodynamic conditions are clearly
very, very different and the concentrations of materials are clearly
very different. But there are some interesting analogies.
MR. LEVINSON: I'd like to make a couple of comments. One,
I want to emphasize something that Ray said that slid by very quickly,
because it was one of the points I had before, and that is everybody is
talking as though uncertainties were all negative.
In fact, that's not true at all. There is a substantial
number of uncertainties which are positive, that reduce dispersion of
materials, et cetera, and we just have to remember that not all
uncertainties are negative in any sense of the word.
MR. GARRICK: What you're saying is that an uncertainty
distribution has a negative side and a positive side.
MR. LEVINSON: Absolutely, absolutely. But we talk about it
as though all uncertainties were bad. As I sit here and listen, I hear
more and more reasons for why the waste issue and the reactor issue
really are very, very different sorts of things. For instance, in the
waste thing, after you start out, the potential risk steadily
deteriorates as stuff decays away.
At a reactor site, the potential risk increases, as over the
life of the reactor, you continually increase the inventory of fission
products on the site. Thing after thing.
Bob showed his dose curves out at one MR or ten MR, it
doesn't make any difference. When you get to the top of the chart, rate
is probably at least as important as dose. Bob has, on his chart, a
thousand rad is certain death, but both his wife and mine, in the last
couple of years, have received significantly more than that in treatment
The dose effect -- now, in a reactor accident, the dose rate
basically, from a prompt criticality, it's an instantaneous thing.
There is no way, in a waste disposal, that anybody is going to get a
high rate of dose. So I just think these things are completely
On history, I want to throw in one comment, since I'm
probably the oldest person here. The NRC may have invented the words
defense-in-depth, but they didn't invent the philosophy. When I joined
the project in 1944, DuPont -- and it wasn't the chemical part of the
company, it was the explosives division of DuPont that was in the
Manhattan Project, and they brought that concept.
It was the first lesson I got when I went to work there.
It's been around a long, long time and I don't know that we're going to
define it or cage it in. It's been a very useful device for designers
and builders, and it's been there a long, long time.
Just one other comment. There was a comment by Bob Budnitz
about U.S. policy for shutting down reactors without containment.
Clearly, that's not a technical based issue at all. But the Soviets
have very, very limited -- now, they have more because we've given it to
them, but they had very, very limited ability to do analysis. I
probably know about as much about it as anybody in this room, since I
spent eight years on the board of directors of the Soviet Nuclear
They did do an analysis in regard to shutting down the RBMKs
at Chernobyl and in a very basic way, in one of the discussions I had
with them, they said maybe our risk of duplication of the Chernobyl
accident is ten-to-the-minus-third, and I said is that acceptable to
you, and they said, wait, we haven't finished telling you the analysis.
If we duplicate the Chernobyl accident, we'll kill 30-some
people. If we shut it down tomorrow, probably ten times that many will
die this first winter. And in this country, we have the luxury of being
able to say you can shut down a reactor without major consequences. In
other parts of the world, that's not the case at all.
Their analysis -- it isn't that they have different values
for what's an acceptable number; they have other considerations.
MR. MURLEY: Tom, could I ask a question that occurred to me
about your concept of allocation? I guess I have different reaction, if
you want to impose it as a requirement or if it's a target.
If it's a kind of aiming goal or target, I think that's a
very good concept. But if you're suggesting that it become embedded in
regulations or something, I have a different reaction about it.
MR. KRESS: And I'm sorry to tell you I had the second, the
latter. The reason I have that is I think in a risk-based regulatory --
risk-informed regulatory system, you can no longer have targets for
individual plants. You have to have risk acceptance criteria for
If you have to have those, then they have to be part of the
regulation. So I really did mean the latter, which I know gives you
With that, I think this is a good time for us to break for
lunch until 1:00, at which time we will hear some interesting comments
from the staff. We're recessed until 1:00.
[Whereupon, at 12:05 p.m., the meeting was recessed, to
reconvene at 1:00 p.m., this same day.]. A F T E R N O O N S E S S I O N [1:00 p.m.]
MR. KRESS: The meeting will come back to order, please.
Before we get started, there's just a very minor change in
the agenda I'd like to point out to people. We were up to item five on
the agenda, which was NRC staff presentations by Gary Holahan and Tom
Instead, we're going to interchange that with item six,
because of some problems, and we're going to have the NRC staff
presentations on the defense-in-depth in high level waste first, and
then move to the defense-in-depth in reactor regulation.
So with that, I will turn the floor over to John Greeves.
MR. GREEVES: My name is John Greeves. I'm Director of the
Division of Waste Management in the Office of Nuclear Materials Safety
and Safeguards. Mr. Chairman, let me thank you for making a schedule
change. Norm Eisenberg, the principal brief, is coming down with
something. He's been coming down with it for days and I think he's sort
of running out of energy. So we thank you for your discretion in
leaving the schedule a little bit.
We also apologize to the audience for moving the time around
a little bit, but for the sake of Norm being able to deliver his
presentation, I think it was the best thing to do.
Again, I am the Director of the Division of Waste
Management. I have spent a fair amount of time interacting with the
Advisory Committee on Nuclear Waste. So obviously this is a time for us
to comment and bring some of our ideas to the process.
I appreciate the difficulty which people were addressing
this issue this morning. Defense-in-depth for materials and waste
licensing actions presents a number of challenges, and you bumped into a
number of those this morning.
Unlike reactors, we have the full spectrum of activities
within NMSS, from exempt sources, which you discussed this morning,
medical activities, sealed sources, fuel fabrication facilities,
transportation, low level waste, high level waste.
It's really a family of different types of licensing
activities. So I think a lot of that was brought out this morning. I
was pleased to see that. I was also heartened by some of the views
expressed. I can tell you there's a number of views within the staff on
these issues, also.
The topics, depending on what type of a licensing activity
you're talking about, have different time spans, have different radio
activity, have different human action, have different criteria, and have
different rates. You touched on all that this morning.
I would like to just punctuate that the staff certainly
looks at the Commission policy statement on risk-informed
performance-based regulation, and I think it's probably in your package
and it has a definition on defense-in-depth, and the staff, in its
efforts, is looking to make sure we stay consistent with that particular
policy statement. It's on the web and is available to people.
As I said, Norm Eisenberg, Dr. Eisenberg is walking this
way. I'll try and not get too close to him. Norm is going to do the
principal presentation. He's going to try and set the context for all
the materials types of activities and a couple of things about Norm.
One, this may be your last chance. He's retiring this
month. He's moving on. The second thing is I think he's a
defense-in-depth expert. This is a gentleman that lives defense-
in-depth. When he gets up, you will notice that he has belts and
suspenders. I've heard statements that people thought they were the
best at certain things. Norm lives this issue.
The second presentation will be by Christiana Lui, to my
left, and that's more focused on Yucca Mountain specifically. I will
have some wrap-up statements regarding that.
As I said, we keep in mind the Commission policy statement
and what we are expressing are our preliminary considerations on a
number of these issues.
With that, I'm going to stop and ask Norm to go through what
I think is a thoughtful presentation. I think it's a bit
thought-provoking, as some of you put forth earlier.
MR. BERNERO: Do you have slides handed out?
MR. GREEVES: There are slides, should be. Norm, you
concentrate on the presentation. We'll get the slides to Bob.
With that, Norm, take over.
MR. EISENBERG: Thank you. I appreciate the subcommittee
letting me go ahead and do this. I am feeling under the weather and I
feel confident that if I start to become incoherent, nobody will notice.
They'll just figure it's me acting normally.
I should say that I'm going to talk about a provisional NMSS
perspective on defense-in-depth for risk-informed performance- based
regulation. These are some staff ideas that have been circulating
around and a lot of them were sharpened by considering the case for high
level waste regulation.
So you have to understand that these are provisional ideas
and they are subject to change.
So what I intend to talk about are what are some of the
motivations for defense-in-depth in NMSS; what are some of the current
things that are causing us to focus on it; what is it, which, of course,
we've heard a lot of discussion about that this morning; how does
defense-in-depth differ from margin and other safety concepts, which I
think is a very important issue; what are some provisional conclusions;
what are some things that we have to determine if we're going to follow
this path; and then I'd like to make a summary.
So NMSS has been engaged in a number of activities that
prompt a focus on defense-in-depth and a risk-informed performance-based
One of the first things is SECY 99-100, which was approved
by the Commission, which is an activity to develop a framework for
materials regulation similar to the framework for reactor regulation
that was developed by the Offices of Research and Nuclear Reactor
Regulation for risk-informing selected NMSS activities.
So this certainly has brought the subject up, certainly the
consideration of refining the approach on high level waste regulation,
as indicated in the proposed Part 63, is another area where
defense-in-depth needed to be considered, and we got a fair number of
public comments on that aspect of the proposed rule.
There are other activities in specific areas, interim spent
fuel storage facilities are being risk-informed. We have ISAs, which is
a type of risk assessment for fuel cycle facilities, and we are
risk-informing the transportation regulation. So there is a lot of
current interest in this.
Let me just say that the performance-based aspect of risk-
informed performance-based regulation places an emphasis on the overall
system performance and the risk-informed aspect considers the
uncertainties and the sources of those uncertainties.
All right. So what's the regulatory environment in NMSS
that we have to deal with? First of all, we have a lot of diversity.
We have a wide range of licensees and systems regulated. They have
varying degrees of complexity, everything from gaseous diffusion plants,
which are complex, to smoke detectors, which are not.
Different systems have different degrees of human
interaction or are dominated by human interaction. We have certainly
different levels of hazard. Some things are not very hazardous at all.
This gives rise to general licenses. Other things are, frankly,
There's diverse capabilities among our licensees for being
able to do analyses of any kind and especially risk analyses, and
there's many different tradeoffs in the need for risk-informed
regulation, the benefits and the costs in different areas that we
We also need to consider, if you will, the taxonomy of the
risks, and Bob Bernero alluded to this earlier, that we have individual
risk to workers and we have the individual risk to members of the
public. We have normal risks and accident risks. We have perceived
risks and actual risks and we have a variety of initiators, mechanical
failures, external events and human error are some of the things.
MR. APOSTOLAKIS: Why do you have perceived risk?
MR. EISENBERG: Because we have to consider the
communication with the public and even though the actual risk in
quantitative terms may be small, the public reaction may be great. So
there will be a response. So we have to consider not just the actual
risks, but, to some degree, the perception of risk by the public, by
policy-makers, and others.
MR. APOSTOLAKIS: But I realize that communication is
important and so on, but surely you're not implying that you will take
actions based on perceived risk rather than actual, as actual meaning
technical. We are not regulating based on perceived risk, are we?
MR. EISENBERG: The agency may have to respond to some
things with an effort which is not in proportion to the actual risk
MR. APOSTOLAKIS: That I agree with and I think, in fact,
the cornerstones that we have on the reactor side are the result of
MR. EISENBERG: I'm just trying to lay this out as the
environment in which we work. Now, how we actually treat it is another
issue, but it is a factor and it does influence what goes on.
MR. APOSTOLAKIS: I agree that it is a factor.
MR. EISENBERG: Well, I'm glad you agree with me. So kind
of moving to the next step, what are the factors for defense-in- depth
in NMSS, what's the current status?
Well, it's the nature of the licensees and the activities
regulated. We have to recognize that NMSS, by and large, regulates
systems with less hazard than nuclear power reactors. NMSS regulations
are a mix of performance-based and risk-informed regulations versus
prescriptive and deterministic regulations.
This is a little bit different, from my understanding of the
reactor side, where things have been dominantly a deterministic
approach. And for some NMSS licensed activities, the hazard does not
warrant a very strong preventative measure of any type, whatever they
are, performance-based or prescriptive or anything. The risks are too
low. Once again, general licenses are not worth very much concern.
Okay. So what's the NMSS safety philosophy? Well, our
strategic plan says that we want reasonable assurance of protecting
public health and safety, common defense and security, and the
environment. Some concepts that assist in achieving defense-in-depth in
this context are safety margin, diversity, redundancy, no single point
of failure, and quality assurance. There is a whole spectrum of things
we do to try to achieve reasonable assurance.
And in this context, defense-in-depth is a component of a
risk management strategy. This does not imply that we do risk
management, all the risk management that a licensee might want to do.
They have other reason to do risk management, but we are obligated to do
risk management in the public health and safety context.
MR. KRESS: When you say risk management, what exactly do
you mean there, Norm?
MR. EISENBERG: In other words, putting forward a structure
of regulations makes certain things less likely and other things more
likely and it is a way of determining what the risks are and how large
they might be allowed to become.
If you take the Kaplan-Garrick definition of risk as the
risk tripled, then regulations provide one constraint on the risk,
meaning that whole aggregate of points.
MR. KRESS: I think I know what you mean now.
MR. EISENBERG: Okay. All right. So if we're going to use
defense-in-depth to help achieve our top level goals of public health
and safety, what is it? Well, this is what was taken, and I forget who
threw it up this morning, but this is from the Commission white paper on
risk-informed performance-based regulation, and this is a paraphrase of
the two key features for defense-in-depth, which are, one, safety is not
wholly dependent on any single element of the system and, two,
incorporation of defense-in-depth into a system produces a facility that
has greater tolerance of failures and external challenges.
MR. KRESS: That's a pretty loose definition.
MR. APOSTOLAKIS: It's, in fact, not a definition.
MR. GREEVES: This is right out of the Commission paper.
MR. APOSTOLAKIS: We realize that.
MR. KRESS: We realize that. Thank you.
MR. APOSTOLAKIS: I thought our comment at the time was that
this is still evolving.
MR. GREEVES: This is what the staff is looking at in terms
of guiding its efforts and being consistent with the Commission paper.
MR. EISENBERG: We took this as one of our starting points.
MR. BERNERO: This is the same thing I put up. This is just
a paraphrase of it.
MR. APOSTOLAKIS: It's what? I'm sorry.
MR. BERNERO: It's the paragraph I put up. The paragraph
that I put up on the screen, this is a paraphrase of it. It's one of
the attempts at defining defense-in-depth. You've got a whole book full
MR. EISENBERG: And here is the whole statement, which I
think -- okay. Well --
MR. GARRICK: I think if you put it in the context we were
discussing this morning as a way of doing business, as a way of how we
provide protection, it fits in that scheme.
MR. EISENBERG: So then the question is how do you do
defense-in-depth in a risk-informed performance-based context. Things
change when you get into a risk-informed performance-based context,
rather than a prescriptive deterministic context. This, I thought, was
stated very nicely in this paper by Sorenson, et al, in which there was
the structuralist and rationalist approach.
So this is, once again, a paraphrase and may not be complete
enough to satisfy everybody in the audience, but basically the
structuralist approach maintains that the need for and extent of
defense-in-depth is related to the system, structure. Many
manifestations are based on the novitant perspectives that were current
at the time that the systems were developed or they were first licensed
and some manifestations have an ad hoc basis.
The rationalist approach articulates a philosophy that says
defense-in-depth should be related to the residual uncertainties in the
system and the rationalist approach is just beginning to be adopted in
this risk-informed, performance-based environment.
And we have taken the structuralist -- I'm sorry -- the
rationalist approach as appropriate for risk-informed performance-based
regulation. But the question is how do you implement it and what are
those uncertainties that you need to address.
MR. APOSTOLAKIS: What do you mean by residual
MR. EISENBERG: Yes.
MR. APOSTOLAKIS: Okay. There is something that --
MR. GREEVES: I'm going to talk more about this.
MR. APOSTOLAKIS: Is there something wrong with the word
unquantified or why are you avoiding it?
MR. GARRICK: Don't be so sensitive, George.
MR. APOSTOLAKIS: Residual is different, because some of the
residual uncertainties have been quantified.
MR. EISENBERG: Remember, what we're assuming here is that
you have a risk-informed performance-based approach. So you've already
folded into your compliance demonstration -- this is very much the case
with Part 63. You've already folded into your compliance demonstration
MR. APOSTOLAKIS: I understand.
MR. EISENBERG: -- consideration of the uncertainties that
you have quantified. They are in there.
MR. APOSTOLAKIS: Right.
MR. EISENBERG: And whatever the criterion is, and for Part
63, it's that the peak of the mean dose be less than 25 millirem, as
long as you meet that, you're okay.
MR. APOSTOLAKIS: But what I'm saying is that after I have
implemented the risk-informed system, yes, I will tolerate certain --
some uncertainty that things will go the wrong way. But that doesn't
mean I'm going to invoke defense-in-depth to handle those, because those
I have quantified.
It's the things that I have not included in my analysis. So
the word residual perhaps is not so fortunate.
MR. GREEVES: He's got some slides that are going to touch
on your issue.
MR. APOSTOLAKIS: I think conceptually we agree.
MR. GREEVES: I think he's going to hit another button here
MR. EISENBERG: Just briefly. So what are the uncertainties
that we consider in these safety assessments, and there's
MR. BUDNITZ: Regulatory.
MR. EISENBERG: Well, there is that differentiation, but
there is also, for those of us that are doing the pragmatic, there's
parameter of data uncertainty, there's model uncertainty, there's
scenario uncertainties, which, for a lot of waste work, involves the
exposure scenario as opposed to some physical scenario, and, also,
programmatic factors; the safety culture, for example.
So this is one cut at uncertainty.
MR. GARRICK: And on way you could look at that, Norm, is I
might even view scenario uncertainty as an integral part of the modeling
uncertainty, given that the scenarios are usually a fundamental part of
the modeling process.
MR. EISENBERG: It's the model of the world or the model of
MR. GARRICK: And the programmatic factors, like QA, those
are there primarily because we don't normally address them explicitly.
In other words, it's not that they couldn't be, it's just that we don't.
MR. APOSTOLAKIS: In fact, the last three, I call them
modeling uncertainty, but if it makes you happy, that's fine.
MR. GARRICK: Well, we agree.
MR. APOSTOLAKIS: We don't want to make Norm unhappy. Not
MR. EISENBERG: Okay. So now, if we get back to the
residual uncertainties or the unquantified uncertainties, I would
suggest that there may be two types. The first type is if you have the
best available risk assessment, if you do the best possible job you
could do, there are still unquantified uncertainties and it's because
human knowledge is finite and you just can't put everything in there.
You don't know everything.
So that's one type of uncertainty. But there's another type
of uncertainty and that's got to do with there's practical realities and
we can't always get the best available risk assessment. Very often, in
the real world, we have to deal with a risk assessment that was done.
It may not be the best available one. There may be significant flaws.
And we also have to consider, in those cases, that there are
unquantified or residual uncertainties.
MR. BUDNITZ: Norm, as a distinction here, in the first one,
you characterize that you did the best you could. You said the reason
why it's not better still is because the state of knowledge is
incomplete. Now, that's epistemic.
I want to argue to you that there are also aliatory
uncertainties that you can't know well.
MR. APOSTOLAKIS: Like what?
MR. BUDNITZ: Like, for example, suppose you would really
like to characterize the environment below the repository horizon, but
above the saturated zone at Yucca Mountain down to the one meter scale,
but, frankly, we can't. So there is a variability naturally in the
system which is going to cause uncertainty in your performance
assessment, and that is certainly aliatory and not epistemic.
So I think that that's incomplete, as written, unless you
acknowledge that this isn't only the state of knowledge. Some of it has
to do with variability in the natural world, which we can't characterize
MR. EISENBERG: I don't want to get into a semantic
MR. APOSTOLAKIS: We understand what you're saying, though.
MR. EISENBERG: And you can --
MR. BUDNITZ: But it's a crucial conceptual point.
MR. EISENBERG: But some people would argue that all
uncertainty is --
MR. BUDNITZ: We've been there.
MR. EISENBERG: -- epistemic. It's not worth talking about.
I mean, some people would argue what you're talking about is the
inability to characterize an aliatory uncertainty.
MR. APOSTOLAKIS: But it's not worth talking about it today.
MR. EISENBERG: Some other time.
MR. BUDNITZ: Except that when you define defense-in-depth,
you need to understand that distinction, I insist.
MR. APOSTOLAKIS: So the second one then would be something
like the IPEs.
MR. EISENBERG: Then I thought I would go into a little
further detail on what these things are, what are the limitations on
knowledge. Well, you may not have included all the failure modes
because you may not know them all and you haven't had enough experience
to learn them all.
You may not have included all the phenomena for the same
reason. The range of variability in the system parameters may be
under-estimated or biased, and this happens not infrequently that people
make an estimate, take data, and their uncertainty increases.
Well, it doesn't mean that the uncertainty increases. It
means that their original estimate of uncertainty was an under-
estimate. Probabilities and consequences for rare events are based on
sparse or non-existent data. Models can't be validated. For the waste
business, we cannot wait 10,000 years to see if our predictions are
Although the systematic analyses methods can give great
insights on how a new system might perform, some problems only come to
light with experience. In other words, the state of knowledge is
evolving. I think that is the bottom line, for one type of uncertainty.
And there is a similar litany for the other kind. Why are
these risk analyses as -- and this includes performance analysis -- why
aren't they as good as they could be. Well, not all failure modes are
included because of limitations on time and resources, because the
people that try to enumerate everything didn't do it right, because not
all the phenomena were included because it would cost too much to model
everything in that detail, because in some cases, only certain kinds of
uncertainty are explicitly represented in the risk assessment.
Parameter uncertainty may or may not be propagated in the
consequence models. Some people would use point estimates. Model
uncertainty may or may not be represented. Probabilities of varies
scenarios and the uncertainty in those probabilities may or may not be
included, and not all the uncertainties that could be quantified have
MR. APOSTOLAKIS: Where are you going with this?
MR. EISENBERG: I'm trying to lay a groundwork that if you
just look at the results of risk assessment and compare it to a safety
goal, that there are uncertainties that you haven't considered.
MR. APOSTOLAKIS: But there is a difference between somebody
saying I will not propagate the parameter uncertainty and somebody
saying I will not do model uncertainty calculations. I will be
extremely hostile to the first guy and very sympathetic to the second,
because it's inexcusable not to propagate parameter uncertainty in
reactors, at least. In your case, it's expensive, but you have other
means to do it.
MR. EISENBERG: But suppose the model uncertainties are the
thing that dominates the result.
MR. APOSTOLAKIS: I understand that, but -- of course. Of
course, model -- but, I mean, just to say real life tells us that some
people don't do parameter uncertainty propagation, I don't know where
that leads us, because that is not something that you can tolerate these
MR. GARRICK: I think the other issue here that is a little
bit troublesome in this regard is this implies that there is an
alternative and if there is an alternative, why doesn't it become a part
of the risk assessment. That's something I'm always wrestling with.
MR. GREEVES: Let me ask you to keep in mind that as Norm
goes through this, this represents our whole program. It's not in Yucca
Mountain and it's not reactors. I think that some people can't afford
to carry these things so far and appropriately so.
So Norm's presentation was trying to give you a spectrum
across the problem that NMSS has.
MR. GARRICK: We'll let him continue.
MR. GREEVES: Okay.
MR. EISENBERG: I was trying to make the point that there
appears to be a case for doing something beyond merely demonstrating
that you meet the risk goal. So before I talk some more about
defense-in-depth, I'd like to try to differentiate between
defense-in-depth and margin, which I think is an important concept, and
I will see how much controversy this raises.
If you will, margin is the cushion between the required
performance of a system and the anticipated or predicted performance.
Defense-in-depth, if you take the quasi definition from the Commission
white paper, is the characteristic of the system not to rely on any
single element of the system and to be more robust to challenges.
Margin describes the expected performance of a system versus
the safety limit. Defense-in-depth describes the ability of the system
to compensate for unanticipated performance results from limitations on
For example, increasing the margin in a system that relies
on a single component doesn't necessarily increase defense-in- depth.
You're still relying on a single component. Defense-in- depth provides
that if any component under-performs, the rest of the system has enough
good qualities in it that it can compensate and provide that the
consequences are not unacceptable.
In going through this briefing for different audiences, some
of the other things that have been suggested is that defense-in- depth
is like a safety net. If you're walking on a high wire and you fall,
the safety net does not assure that you get to the other side. But it
means that you may not get killed. So this can be a good quality of the
The same with seat belts and air bags. Neither one of them
keep you from getting into an automobile accident, but they both may
prevent -- they put a lid on the consequences.
So if I can follow this -- you're shaking your head, George.
MR. APOSTOLAKIS: Finish, and I will tell you why.
MR. BUDNITZ: He wants you to quantify those differences.
MR. EISENBERG: This is an example where there's two systems
and we're assuming that components A, B and C, on the left-hand one, are
diverse and they don't have common cause failures, and they both meet
the same risk goal, but the one on the left has the quality that if any
one component fails to perform as expected, you could still meet the
ten-to-the-minus-four risk goal.
On the system on the right, if that one component is off,
you may have had it.
MR. APOSTOLAKIS: But this is a very misleading example,
Norm. Where are the uncertainties in these numbers? You can't present
an example like this on the basis of point estimates. I would say that
the system on the left, if it's an engineered system, will have smaller
uncertainty about the ten-to-the-minus- six.
So it may be preferable that way.
MR. KRESS: Or it may not.
MR. APOSTOLAKIS: Or it may not. It could be. If we take
the vessel --
MR. KRESS: And you might want to elect it because it --
MR. APOSTOLAKIS: So giving examples like this on the basis
of point estimates doesn't really help.
MR. EISENBERG: Well, what is it that you're shooting for,
and when you say that the uncertainties on the left may be smaller,
you're talking about the quantified uncertainties.
MR. APOSTOLAKIS: Yes.
MR. EISENBERG: And I thought I had made it clear that I was
talking about the unquantified or the residual uncertainties.
MR. APOSTOLAKIS: But even for the original uncertainties, I
would expect them to be smaller on the left.
MR. EISENBERG: Why?
MR. APOSTOLAKIS: Because for systems, components that are
at the ten-to-the-minus-two, in the ten-to-the-minus-two range, I
wouldn't expect the residual uncertainties of the unquantified to be
Now, you might say but if you put them together, there might
be something. Still, I wouldn't expect the probability of a dependency
that would defeat three components to be so significant as to overwhelm
the probability that one component that I wanted to be so reliable at
the ten-to-the-minus-six level, you know, the uncertainties are
The whole issue of defense-in-depth is an issue of
uncertainty in the frequencies, not to the point values. If we don't
accept that, then defense-in-depth doesn't make any sense or it will be
a principal forever.
MR. EISENBERG: I guess I don't understand how you would
fold in to this consideration the unquantified uncertainties.
MR. APOSTOLAKIS: Because if I had to have the discussion I
mentioned this morning, focusing on the unquantified uncertainties, I
would have a bunch of experts arguing why, how can a system with three
components, a particular way it's configured, first of all, that must be
an "and" gate, not an "or" gate.
MR. EISENBERG: Yes.
MR. APOSTOLAKIS: And/or, what does it matter, right? It's
an "and" gate. They would have to focus on these -- on the failure
modes of a three-component system that would defeat all three of them at
the same time and express whatever uncertainty they have about those,
and it seems to me that is something that -- that's the value of
By spreading it over three components, this residual risk is
smaller than on the right, where you have one. Think about all - - if
you read the documents from the agency over the last 40 years, I think
that's the running philosophy and I had about ten quotations from SECY
98-225, where the issue of confidence, uncertainty comes up every other
Anyway, that's my view and we can continue.
MR. EISENBERG: I think you're agreeing with me.
MR. APOSTOLAKIS: I won't do it on the basis of point
values, because my basic thesis is that defense-in-depth deals with the
uncertainties in these probabilities, frequencies.
MR. EISENBERG: One way of thinking about defense-in-depth
in the NMSS context is there appear to be two things that you want to be
concerned about. One is the hazard level and the other is the
uncertainty in the performance of the safety system. Here, again, I'm
talking about the residual uncertainty or the unquantified uncertainty.
This is not necessarily related to the behavior of the
system as modeled. It's related to the experience with the system,
whether, in fact, it ever has been built and operated or tested. So
there's a qualitative scale. This is not intended to be quantitative.
There is a qualitative scale in the Y axis that relates to the degree of
There is a qualitative scale on the horizontal axis that
relates to the hazard. Small hazard, you don't need much
defense-in-depth because the consequences are not great. High hazard,
you need more defense-in-depth. So this kind of outlines three bands of
degrees of defense-in-depth and depending upon where you fall on a chart
like this or, in practice, the way we have decided to regulate these
determines how much defense-in- depth you have in each area.
But this might be a semi-quantitative, but rational approach
to deciding how much defense-in-depth is needed based on these two
Now, there may be other qualities that are important in
making those decisions, also. This is a suggestion of how we might
approach it on, let's say, an NMSS-wide basis.
MR. APOSTOLAKIS: I like it. I like it a lot as a first
step and I think pictorially it shows -- I mean, I would translate that,
again, to uncertainty language. What you're saying is that if the
hazard is high, I really have an interest in the consequences. If it's
small, I probably don't care. If it's high, I have an interest.
And then on the vertical scale, you have put it very well.
If I have data and experience, in my language, there is no residual
uncertainty, there is no need for defense-in-depth.
So this is great. And as you move up, you hit a brick wall.
MR. KRESS: I'm wondering why you chose to stair-step this
particular thing instead of straight lines.
MR. EISENBERG: I think it's easier with the graphics
MR. KRESS: Okay.
MR. APOSTOLAKIS: I must say, though, that your presentation
up to now probably has nothing to do with this.
MR. EISENBERG: We thought it did.
MR. APOSTOLAKIS: I think you could have started with this.
That's not a criticism.
MR. GREEVES: I think this kind of conveys the spectrum of
issues that challenge NMSS. It's multiple licenses and we've got we've
got to think in this context.
MR. APOSTOLAKIS: But, see, the problem I had with your
earlier viewgraphs is -- and I don't -- I suspect you didn't mean that,
but I don't think we should regulate taking into account the fact that
people don't like to do a few things, like propagating parameter
On the other hand, you may have a problem on your hands with
the medical uses, all this, and where do you draw the line? I don't
know myself. When do you say, no, you have to do this? Otherwise, we
will do such and such a thing to you.
And I have seen nothing in this diagram that is based on
that. That's what I meant, that it's independent of what you presented
I take the vertical axis as meaning it's an objective axis.
It says it has never been analyzed. That's a statement of fact.
Analysis are confirmed by data. That's a statement of fact. It has
nothing to do with the choices that the licensee makes.
MR. EISENBERG: This is choices for us. This is choices for
us and the preceding material, I think, made two points. One is that
it's the unquantified or the residual uncertainty that should have an
effect on how much defense-in-depth you need and, also, that what you're
really concerned with is not what the risk is. It's with the hazard
level, because the potential there is that if you're relying heavily on
a single element of your system, if you didn't do something right and
something goes wrong, you can be in trouble.
So it's the hazard and the residual uncertainty that you
really want to think about, not necessarily risk. Risk we covered
because we already said we were operating in a risk- informed
MR. GARRICK: You want to be a little careful with pushing
this too far, because if you're concerned about dose, let us say, and
you have ten-to-the-ninth curies of fission products in one mode versus
another mode, the problems are grossly different.
In the case of a reactor, where you have lots of stored
energy and you have lots of mechanisms to enhance the distribution of
this material, that's very much different than having ten-to-the-ninth
curies in an unstored energy environment.
So you really have to be careful about drawing too many
conclusions about risk from these kind of diagrams.
MR. EISENBERG: I agree with you, and you also do not want
to use this as an open-ended invitation to require more and more things.
You don't want to imagine totally impossible or extremely unrealistic
MR. APOSTOLAKIS: I think this is a good communication tool,
that's all it is. It really conveys the idea. I don't see how you can
make this practical. You're going to tell us later, right?
MR. EISENBERG: Yucca Mountain is somewhere on the graph. I
don't think it's got as much hazard as a power reactor, but I don't
think we have as much experience with it as we do for the power
reactors. We don't have it built and tested yet.
Christiana is going to answer your question, because she is
going to tell you how --
MR. APOSTOLAKIS: You're doing a pretty good job yourself of
that. Don't be so defensive.
MR. EISENBERG: But in terms of how it's being implemented,
we're still working on it and maybe the first thing out of the box is
Yucca Mountain and we haven't gotten all the way there on that yet
Remember, the comment period is closed. We're working on
developing the position. We haven't gotten it up to the Commission yet.
So what are the conclusions about defense-in-depth, some
provisional conclusions? Well, it's related to, but different from
other safety concepts like margin. It's not equivalent to meeting a
safety goal or the margin to be associated with meeting the goal. It
can be implemented in a risk-informed performance-based context as a
system requirement rather than as a set of subsystem requirements.
So that what we would suggest is that you can look at the
uncertainty, the residual uncertainty related to any particular barrier
in your system or any particular feature of your system and demand a
degree of defense-in-depth that is proportional to the uncertainty.
More uncertainty, you want more defense-in- depth. And all this is
leavened by the amount of hazard.
MR. APOSTOLAKIS: Now, that's an interesting thought. You
say you would look at each element and the residual uncertainty and do
this. How about if I take another approach? I look at each element, I
look at the residual uncertainty in each one. But then I use a
convolution there to find the residual uncertainty regarding the
performance of the whole system and then I impose defense-in-depth.
What's wrong with that? Instead of doing it at each
MR. EISENBERG: Let me be clear. If you do it on an
element-by-element basis, it's all pointing at the ultimate risk goal.
It's all pointing to the performance objective.
MR. GARRICK: So your answer is you agree with it.
MR. APOSTOLAKIS: You agree with me.
MR. EISENBERG: I think we agree again.
MR. APOSTOLAKIS: Or it could be a combination of the two.
MR. KRESS: Let me sort of rephrase what I heard. I've
heard that more the residual uncertainty, and George has qualified
residual to mean unquantified, the more the defense-in- depth you need
and then George says you use defense-in-depth where you have
unquantified uncertainties, so you don't know what the meaning of the
word more is, and I keep saying you do have to quantify it.
I'm a little confused. What are we talking about here?
MR. APOSTOLAKIS: Unquantified in the sense that I hadn't
put down a probability distribution. But there is something, in my
mind, I mean --
MR. KRESS: You mean, it's big or medium or small?
MR. APOSTOLAKIS: Yes. I could say --
MR. KRESS: Isn't that quantified? See, I'm saying you can
quantify it to some extent.
MR. APOSTOLAKIS: To some extent, I agree. Yes. You're
MR. GARRICK: And I agree with you, Tom. It's a very
abstract concept. In fact, I still struggle with what we mean by
unquantified or residual uncertainty and if we can handle it by some
other means, why can't we fold it into the basic parameters.
MR. APOSTOLAKIS: We could. We could. We could.
MR. BUDNITZ: I don't understand why, George, it's the
unquantified uncertainty and only that that you're emphasizing. I can
conjure up a system where it's a quantified, but large aliatory
uncertainty and you invoke defense-in-depth to find a way to do it
anyway that's safe enough.
MR. APOSTOLAKIS: I would say, in that case, I would use the
uncertainty diversity and so on to manage that uncertainty.
MR. BUDNITZ: In other words, aliatory is something that's
random in nature.
MR. APOSTOLAKIS: That's fine.
MR. BUDNITZ: But large, but we don't know how to control
it. So we find another way using defense-in-depth. But in that sense
MR. APOSTOLAKIS: But it's not defense-in-depth anymore in
the sense that it's not arbitrary. If I postulate a barrier, I can
MR. BUDNITZ: Defense-in-depth isn't arbitrary here. He
said defense-in-depth involves -- we're now going back to the white
paper -- it involves assuring that there's -- you're not relying only on
MR. APOSTOLAKIS: But that's arbitrary.
MR. BUDNITZ: Well, wait. Whatever you say, however they
defined it, I insist that I think it is not only the unquantified
uncertainty, by any means, especially in some of their systems, where
they may have a very large -- by the way, aliatory, maybe they have 800
licensees and they're all different in the arena of some little thing
and in order to have one rule for them, they may have to do it another
way, with the defense-in-depth idea, but maybe two barriers or
something, rather than -- so that might be a variability in nature,
because all the hospitals are different or something.
MR. APOSTOLAKIS: Let me tell you --
MR. BUDNITZ: It's more than unquantified uncertainty, is my
MR. EISENBERG: But remember, this is predicated on meeting
already the risk-informed performance-based goals.
MR. BUDNITZ: I understand that.
MR. EISENBERG: Your aliatory uncertainties, if you have
included them, have already been taken care of. You've already arrived
at a satisfactory performance of the system.
MR. BUDNITZ: I understand.
MR. APOSTOLAKIS: I want to give an example, John Garrick,
what is an unquantified uncertainty. If there is a fire in a nuclear
plant, we have now a methodology that calculates, to some extent anyway,
but it calculates the probabilities of failure of cables and so on due
We know that the fire creates smoke and we know smoke is
hazardous. Yet, right now, we are not quantifying -- this is not part
of my risk assessment. So I can say now, okay, that's not part of your
risk assessment, defense-in-depth, help. So I want you to have barriers
between compartments so that smoke doesn't propagate, I want you to have
smoke detectors, I want the people to have masks and oxygen and this and
So I'm giving you a set of measures and you say, fine, I'll
implement them. This is a traditional way of regulating defense-
in-depth. Then tomorrow somebody does a calculation and he includes
smoke into this, into the fire risk assessment. Now I can see what the
impact on the frequencies of failure, for example, of core damage or
whatever is of having those barriers or having the oxygen masks and so
on, and I may very well decide that some of them are not needed.
So that's what I mean by unquantified, that you invoke then
the principle of traditional engineering and you say then put a few
barriers there that make sense.
In this particular case, I happen to believe that given
sufficient time and will, we can include it in the fire risk assessment.
It's not something -- it's not like safety culture, which is much more
So that's what I mean by -- and then we will just have to do
-- and from the engineering perspective, does this make sense? Yes. To
contain the smoke and make sure that people are not hurt and so on, the
firefighters and so on. So you are invoking a series of measures to
manage this risk, which you have not quantified at this time, and it may
very well turn out in the future that some of these measures were not
the best or were not necessary, they contributed very little, after you
quantified it. It's very good.
MR. EISENBERG: I think we have two problems in our arena.
We have a diverse set of things we regulate. So for each arena, we have
to decide how much defense-in-depth should we have for this particular
set of licensees, how much should we have for the radiographers, how
much should we have for medical licensees.
Then once we decide that, within each system, we have to
decide how do we put in defense-in-depth appropriately to counter the
residual uncertainty. So it's a two-step question.
MR. APOSTOLAKIS: I agree.
MR. EISENBERG: So we think that defense-in-depth can be
used to address these residual uncertainties and we also think that it
should depend on the degree of residual uncertainty and the degree of
But it's not easy. Regulatory life is not easy. So given
this, we still have to decide how to measure the degree of
defense-in-depth, how to measure the degree of uncertainty in the
performance of the safety system, encompassing both quantified and
unquantified uncertainty; how do we measure the potential hazard posed
by a system.
Some of these we've already discussed. How to implement
defense-in-depth when there is different uncertainties in different
parts of the system; how do you use the current state of knowledge to
make reasonable tests for the system to have an appropriate degree of
defense-in-depth when what you're trying to accommodate is imperfect
And then the real killer, how do you explain this to
stakeholders so that we can preserve the flexibility that's inherent in
a risk-informed performance-based approach to defense-in-depth, but also
provide for reasonable assurance of safety. This is not easy.
MR. KRESS: I think this is a good list of issues.
MR. EISENBERG: So in summary, we intend to consider
defense-in-depth in the context of risk-informed performance- based
regulation and a lot of ongoing activities and as part of the continuing
evolution of the risk-informed framework in NMSS.
As a general safety principle, the degree of defense-in-
depth needed to assure safety depends on several factors, including the
degree of residual uncertainty and the degree of hazard. We would like
to implement defense-in-depth as a system requirement, where feasible,
rather than by prescriptive subsystem requirements, and please remember,
NMSS needs flexibility in any overall approach to implementing
defense-in- depth to permit us to appropriately regulate the wide range
of systems and licensees that we have.
MR. APOSTOLAKIS: I think this is very good, Norm. You did
a good job.
MR. EISENBERG: Thank you.
MR. APOSTOLAKIS: Even if I sounded critical. The only
thing that bothers me a little bit is this degree of hazard. I'm sure
there is another way of putting it, but for this stage of development, I
guess it's okay.
I think it has probably to do with the goals, the risk
goals, that the degree of hazards affects the goals, the acceptance
criteria, and then that affects the residual uncertainty. So it's
really only one of the hollow bullets there that come at us.
MR. EISENBERG: I'm not sure I agree.
MR. APOSTOLAKIS: The degree of hazard, how you manage it is
a policy issue and the Commission says I have the quantitative health
objectives. Then trying to quantify now your actual system to compare
with your objectives, you end up with a residual uncertainty which is
driven by the Commission's health objectives.
If the Commission had told me that ten-to-the-minus-two is
the individual risk I will tolerate from nuclear reactors, I will need
to worry about residual uncertainty in nuclear power plants. Right?
The goal is so high that it's irrelevant.
So I think the goal itself is really the driver that
determines the residual uncertainty. But that's a technicality.
MR. EISENBERG: You're tending to look at uncertainties
strictly in terms of uncertainty in frequencies of events of failures.
MR. APOSTOLAKIS: Uncertainty about the occurrence of
MR. EISENBERG: I think that's what I said. But there are a
lot of other ways that the uncertainty can come in.
MR. GARRICK: My concern with the statement, the bullet on
degree of hazard, is a little different. I think that I worry about the
non-linearity between hazard and risk. I wouldn't bank too much on the
degree of hazard being a particularly important factor on this.
MR. APOSTOLAKIS: I think there will be other things driven
by the degree of hazard that will have more direct impact.
MR. KRESS: I would like to see a statement of what is meant
by degree of hazard. I would have interpreted it to mean that if I
didn't have any of the protective systems around this piece of scrap,
whatever it is, the reactor or what, then what is the probability of
producing certain consequences.
If we just laid the fission products in the hole up there,
why, you can come up with it, or if you didn't have any protective
systems around a reactor, you would conclude that the degree of hazard
of the reactor is much, much greater than one of a repository.
I think you can quantify the degree of hazard, if you just
ask yourself what it means. And it would incorporate your comment about
driving forces and mobility and where it can go and that sort of thing.
MR. EISENBERG: One of the problems of just considering the
risk is that the risk is predicated upon things behaving as they have
been modeled, and one of the things you want to get to with
defense-in-depth is what if, what if they do not behave that way.
MR. GARRICK: Of course, you can even take into account that
by the way in which you assign uncertainty to your model parameters.
There is nothing that prevents you from even accounting for residual
risk at the parameter level or at the barrier level by how you assign
your uncertainty, as long as you've got a case for it, as long as you've
got a story behind it. And I would agree with George. That was a good
MR. GREEVES: And I think we'll keep Norm up here.
Christiana, at this point, as I introduced, the challenge that we have
is thinking across all of the NMSS activities and Christiana Lui will
give you some insight of our current thinking in the Yucca Mountain
So Norm will stick around, because I'm sure it's going to
cause some additional discussion. Christiana?
MS. LUI: As Norm is getting his act together. Thank you.
Good afternoon. My name is Christiana Lui and I work in the Division of
Waste Management in the High Level Waste Branch, and we heard a lot --
we heard a lot of interesting discussion this morning and hopefully in
my presentation I will be able to help answer some of the questions and
make some clarifications to some of the issues that have been raised
regarding the high level waste program this morning.
I just want to provide the context of where we are. The
extended public comment period on the proposed Part 63 ended on June 30,
1999. Staff is in the process of analyzing the public comments and
preparing responses to those public comments.
The current schedule is to have the final Part 63 go to
Commission by the end of March this year.
Again, to emphasize that this is still work in progress. So
the objective today is to share our best current thinking with the
committee, and the focus is going to be on the post-closure safety
evaluation, how the multiple barriers requirement is being addressed in
the post-closure safety evaluation.
For pre-closure, the defense-in-depth follows the approach
of prevention, mitigation, and if you want to put emergency planning, a
separate category, but basically it's the same concept as the operating
facilities that you are most -- you are definitely will hear from our
colleagues from NRR and Research in the next two presentations.
I'm going to go from pretty much the very top level and
provide more detail as the progression of the presentation. So we want
to clarify what is the intent of multiple barriers first.
Just a side note that we received approximately 20 sets of
public comments on the issue of multiple barriers during the public
comment period, including Dr. Budnitz's comment asking us to clarify
what we mean by the multiple barrier requirement in Part 63, and we
appreciate your comment.
As both John and Norm have mentioned, the intent of the
multiple barriers is we are going to -- we are using the Commission's
white paper on the risk-informed and performance- based regulation as
the guidance for our approach to clarify multiple barriers requirement.
We also are going to measure at this point. We are
targeting the multiple barrier requirement as an assurance requirement,
and I will say about -- I will provide you more detail on this a little
The known certainties are all captured, appropriately
captured in the performance assessment to demonstrate compliance to an
individual protection standard.
MR. APOSTOLAKIS: Are the model uncertainties also
MS. LUI: Yes. I'm going to talk about that. I'm going to
give you a little bit more detail on that. So just be patient, bear
with me. Thank you.
MR. APOSTOLAKIS: You're asking for the impossible, be
MR. GARRICK: I'll help you, Christiana.
MS. LUI: Okay. And --
MR. APOSTOLAKIS: But wait a minute.
MS. LUI: And the repository system is sufficiently robust
to account for -- maybe imperfect is not the best word here. Maybe
incomplete is a more appropriate word here, the incomplete knowledge.
MR. APOSTOLAKIS: This is the second time that we hear this
today. The first one was from Dr. Budnitz. So it is the community's
view that even without imperfect knowledge and the uncertainties and so
on, we are meeting the goals of the Commission, that Yucca Mountain
meets the goals?
MR. BUDNITZ: We don't know.
MR. APOSTOLAKIS: So what does it mean then, that it's
sufficiently robust or accounts for imperfect knowledge? To do what?
This morning you were more explicit. You said, Bob, that even if I
include those uncertainties, I know that this thing is --
MR. BUDNITZ: I expressed an opinion, but of course, we
don't know, because we don't have a final design or analysis of it. I
was of the opinion that I think it's likely that when the final decision
is put in place and it's analyzed, I think and hope that it will meet
the dose limits in Amergosa with a lot of margin.
MR. GREEVES: In spite of imperfect knowledge.
MR. BUDNITZ: No, not in spite of, taking into account. Not
just in spite of. Taking into account. So that's a prediction, because
I don't know, the final design may have some more difficult analysis
problems than the things I've seen.
So this is still an evolving sort of judgment and I don't
want to preempt even my own final judgment there, but I was just sort of
expressing and I was stipulating that if that's true, then what.
MR. LEVINSON: Well, the slide identifies this as the
intent. It doesn't say they have achieved it.
MR. BUDNITZ: Yes, of course. That's there, yes.
MS. LUI: There will be a lot of discussion. Next slide.
Now I'm going to be a little bit more specific on what are the
considerations of the multiple barriers requirement in Part 63.
I'm going to take you step-by-step here. The reason why the
fourth bullet is in yellow is because that's one particular item not
included in the proposed Part 63, but is being -- but is under
consideration. That as part of the clarifying language for Part 63, we
are intending to add that part to the regulation.
The first thing is to assess all significant and negative
impacts on safety in a compliance demonstration calculation. This
morning -- or what I really mean by that, this morning we have heard
quite a bit about TSPA or that particular terminology being used.
Basically, what we asked DOE to do is in the total system
performance calculation, that they carefully consider all the data
obtained from site characterization program, consider all the applicable
natural analog experimental and field testing information and justify
the models for the total system performance assessment.
In that, they also have to quantify and incorporate the
uncertainty for all the input parameters that go into a calculation.
DOE also needs to take into consideration the alternative conceptual
models that are -- that basically fits all the information that we have
up-to-date, provide that particular description, and provide a
description of what conceptual models they have considered and what they
have chosen to include in the total system performance assessment.
They also have to provide support that a model output is
MR. APOSTOLAKIS: Again, let me play devil's advocate here.
Suppose you hadn't told them that. Don't you think they would have done
all this? This is nothing special about what you are doing. I think
they would have identified the barriers, they would have described and
quantified the capabilities, they would have provided a technical basis.
There is nothing new here.
MS. LUI: But these are the requirements that are under
consideration in Part 63.
MR. APOSTOLAKIS: You mean under consideration that you may
decide not to demand some of this?
MS. LUI: No, because as what John has stated up front, that
we are still in the stage of preparing the final rule package to the
MR. GREEVES: The staff is being a little careful here.
Recognize, we've got a proposed rule on the street. The period of
comment is closed. We're going through a deliberative process, which is
what is in the regulation. I wouldn't make any more than that of it.
MR. APOSTOLAKIS: But there is nothing special to Yucca
Mountain here. I mean, you would do that for any system.
MR. GREEVES: I don't think there is a trick question.
MR. APOSTOLAKIS: Now, this business of wholly dependent.
What does that mean? I can build a --
MR. GARRICK: I hope it doesn't mean that you would
discourage them from providing you a design where a single barrier could
do the job.
MR. APOSTOLAKIS: I think that's what it means.
MR. GREEVES: No, it doesn't mean that.
MS. LUI: No, it's not that.
MR. APOSTOLAKIS: What does it mean?
MR. GARRICK: That would be terrible.
MR. BERNERO: John, there is a statute that says you have to
have multiple barriers. That colored, the fourth bullet could be
interpreted as a way to verify that, but I would think it would be
worded something like unduly dependent, rather than wholly dependent.
MS. LUI: The reason these words are here, they are taken
directly out from the Commission's white paper. We may -- in terms of
the exact language in the rule, that's still being crafted.
MR. BERNERO: But, Christiana, there has to be a finding
somewhere down the road that the statute is satisfied. DOE has to make
that finding in their submittal, and I agree with George, all of these
things are appropriate to a reasonable total system performance
assessment, except that fourth one. That's a ringer in it, because
that's the implementation of multiple barriers, and, by inference, the
implication of defense-in-depth.
MS. LUI: Right.
MR. BERNERO: The statute requires multiple barriers.
MS. LUI: Right.
MR. BERNERO: I would argue that defense-in-depth is a
strategy, not a statutory requirement, and it says don't unduly depend
on one barrier.
But if you could have a state of knowledge and a state of
certainty that could support one barrier doing the job, then you would
have a statutory conflict but not a logical conflict.
MR. BUDNITZ: In fact, let me postulate something that isn't
true. Suppose --
MR. BERNERO: Are you going to tell us a lie?
MR. BUDNITZ: No, no.
MR. BUDNITZ: It is a "suppose" -- suppose DOE came with a
canister design that they had extremely high confidence in they could
back up and everybody agreed the last 20,000 years, all of them, for the
first cracks, just as, by the way, if they asserted that for one year we
would agree, so then I am just supposing.
Now let's suppose they also had a site in which anything
that leaked the travel time was 50,000 years and they had a 10,000 year
requirement. You're home free -- either is wholly dependent, but it's
not because either one can actually be -- you could have them use a
paper bag and still be there and you didn't have to have the earth,
you'd still be there -- and we want to encourage that. Nobody wants to
discourage them from doing as best they can.
MS. LUI: Right.
MR. BUDNITZ: But --
MR. APOSTOLAKIS: So it is a model of language.
MR. BUDNITZ: No, no, but then if that is the case, let me
stick to it -- just pretend -- suppose that was the case. Would the NRC
ask them to do more? I my prepared remarks this morning I asked that
In other words, if you are there --
MR. APOSTOLAKIS: I think the question would be, Bob,
whether you are there. The NRC will ask them -- I mean if you
demonstrate you are there, I don't think the NRC would ask them to do
MR. BUDNITZ: No, no, no, no, no. Wait -- no, no, no. I
want to insist. I ask another question. Let's suppose that the total
system performance assessment they do next year, two years from now, for
the design they are putting together now shows the doses are met by
three orders of magnitude. I insist that as best I can tell the
Department could still flunk on defense-in-depth. It was all one item.
MR. APOSTOLAKIS: I don't know what all one means.
MR. BUDNITZ: Let me describe.
MR. APOSTOLAKIS: I think the paper background, a second
MR. BERNERO: Now let me give you an example. If the
repository was chosen to be in a site that's subject to significantly --
subject to erosion such that the deposited waste could be exposed in the
long range and you did have a gorgeous package, you know, boy, this
package is marvelous, best can in the world, but it could flunk the test
because the erosion would shift you to be wholly dependent on the one as
against unduly dependent on it.
You know, the erosion might be very far-fetched.
MR. APOSTOLAKIS: I understand that.
MR. BERNERO: But your dependence is upon the package.
MR. GARRICK: Well, you have cited a weakness in the
MR. BERNERO: I still argue there is a difference between
defense-in-depth as a strategy or safety philosophy and what the statute
requires the high level waste repository to have, multiple barriers.
MR. APOSTOLAKIS: No, but the point, I agree with John again
that you can't do these things by counting barriers.
MR. BERNERO: Of course.
MR. APOSTOLAKIS: You can't for the same reason that you
can't rank minimal cut sets in a fault tree by counting the number of
events. The probability of failure must play a role. We are not going
to go back 20 years now and I think, you know, I can restate what you
just said, Bob, in terms of uncertainty and probability and then I will
conclude that it relies unduly on one barrier. I can do that.
MR. BUDNITZ: I agree.
MR. APOSTOLAKIS: It all comes down to the probabilities of
failure of pathways and so on, so by saying, you know, multiple barriers
and count them and so on, this is a first step.
MS. LUI: I don't think we are suggesting counting the
MR. APOSTOLAKIS: We were not criticizing you. We are
talking to each other. When we talk to each other --
MS. LUI: Okay.
MR. APOSTOLAKIS: It's best to change viewgraphs.
MS. LUI: Should we go on to the next slide?
MR. APOSTOLAKIS: Yes.
MS. LUI: Okay. On multiple barriers, some of the concepts
we tried to express on these particular slides has actually come out
during the discussion you just had. What I want to make sure is that
because of the uncertainty in the barriers' capabilities based on
current state of knowledge, there are uncertainties in the barriers'
capabilities over 10,000 years and as the regulator why we want to know
is what if all of these barriers do not perform as well as what we
We want to make sure if that kind of situation happens the
public health and safety is still protected, so what we are going to be
aiming at is that the demonstration of multiple barriers is going to
show that the balance of the system has the ability to compensate for
that kind of "what if" situation.
MR. APOSTOLAKIS: Now the "what if" -- are you going to put
any probabilities on the "what if"?
MS. LUI: We do not plan to do that at this point because,
remember, the TSPA is as good knowledge as possible based on the current
state of knowledge. What we are doing here --
MR. APOSTOLAKIS: Sensitivity studies.
MS. LUI: Yes.
MR. APOSTOLAKIS: That is really what you are doing.
MS. LUI: Or it is similar to a stylized calculation like
human intrusion. You really cannot quantify the probability. If you
can, then it should be really part of your TSPA.
MR. APOSTOLAKIS: I would do it in a different way. I would
start with "what if" and let's say that in "what if" Number 5 I do not
protect public health and safety to my satisfaction. Before I do
anything else, I would ask myself whether "what if" Number 5 has a
probability that would really upset all the calculations and the
confidence that I have.
In other words, I would not rely on a "what if" analysis
without addressing the issue of how likely that is.
MR. EISENBERG: But if you are trying to look at your
imperfect state of knowledge, you are speculating about what you don't
MR. APOSTOLAKIS: I am not speculating because --
MR. EISENBERG: Then how do you know --
MR. APOSTOLAKIS: Wait a minute, wait a minute. At some
point you draw the line. I mean there must be some sort of an upper
bound that you can put. I mean it comes down to Tom's point and John's
that you can always give a number or do something, you know? The
problem with "what if" calculations is the same one as defense-in-depth.
There is no control over it.
This committee 20 years ago, 25 years ago, the moment the
Reactor Safety Study hit the streets several members for years took
extra pleasure by taking a few parameters, multiplying by 10 and saying
my god, look what happens to the result, and everybody said yeah, look
at what happens to the result.
The question is can you multiply it by 10? Is that real?
And I think you are going that way. You can start playing games here
that have no bound.
MR. EISENBERG: The key thing here is that the
underperformance would be related to the degree of uncertainty in that
particular barrier, so if you have a very good case, if you have lots of
evidence, then you would underperform it very little. If you don't have
a whole lot of data, if you have a 20,000 year waste package and you
have two months of data, well, maybe we would want to see it
underperformed more, but it is not unbounded speculation and it is not
intended to be unbounded speculation.
MR. BUDNITZ: I have peeked ahead but --
MR. BUDNITZ: -- but it is a fair comment to say that
although I wasn't in Las Vegas in November I read the transcript and
your thinking here is the same as there and that's great because, you
know, it's only been a couple months and I understand what you are
I am still troubled by two things. Unless I peeked ahead
and didn't get it right, you are still asking the Department, the
Applicant, to select the amount of underperformance that they will
analyze, and I think that is not necessarily right.
MR. GREEVES: Well, why don't we move to the next one.
MR. BUDNITZ: Maybe we can go to that.
MR. GREEVES: I am not sure you read that slide right.
MR. BUDNITZ: Maybe I didn't get that one right, but the
second point is on this slide. Go back to this slide. It has to do
with the word "compensate."
The word "compensate," my plain English reading of that
convinces me it is the wrong word. You can't expect that if you
underperform a certain barrier that you would necessarily still meet the
dose limit at Amargosa Valley or maybe you do mean that. It's very
important to understand that.
MS. LUI: Right.
MR. APOSTOLAKIS: What did you say?
MS. LUI: If you look at it carefully, it's not fully
compensated. We are talking about compensate.
MR. BUDNITZ: So let me try to say this. Suppose the dose
limit at Yucca Mountain is "x" millirem per year and the base case
calculation shows one-hundredth of "x" and then they undercompensate
Barrier Number 2, underperform, excuse me, underperform Barrier Number
2, and instead of being .01 of "x," whatever the limit is, it's now 5x.
Do they get a license or don't they?
Now that depends on something that they haven't told us yet.
It's really a crucial point.
MR. APOSTOLAKIS: What is it that you haven't been told?
MR. BUDNITZ: They haven't told us whether or not they are
going to get a license or not.
DR. KRESS: And is that acceptable. You haven't defined an
acceptable performance --
MR. APOSTOLAKIS: Isn't the obvious thing to do to ask
yourself how likely this postulate we made was?
MR. BUDNITZ: That is a piece of it, of course.
MR. APOSTOLAKIS: That is the most important piece.
MR. BUDNITZ: I am not arguing the case, but you see, if in
fact something becomes 5x instead of .01x but "x" is the limit, right?
-- we may all judge that that is sufficiently unlikely that we will give
them the license, right? But they haven't told us, the public, and here
I am a member of the public because I am not under contract to anybody
right now, or certainly they haven't told the Applicant yet, unless I've
peeked ahead and haven't seen it, whether -- what the decision criterion
is and in my remarks I said it has to be fair and it has to be
technically sound and it's very, very important that that be clarified.
MR. APOSTOLAKIS: The weak calculations set a bad precedent
there. Look at the spaghetti curves.
MR. BUDNITZ: Well, we are not arguing the case.
MR. APOSTOLAKIS: All of them are below.
MR. BUDNITZ: You see what I'm saying? So keep going.
MR. GREEVES: I understand what you are saying and you are
not going to be satisfied.
MR. BUDNITZ: I know I am not going to be satisfied and I
want to say that if I was designing the repository and some of the guys
behind me are, and if I was trying to put it together now so that I
could analyze it next year, so I could bring you the thing in the year
after next and I didn't even know whether the design I am contemplating
freezing for this will do this, that is a real problem, that's a real
MR. GARRICK: I think that the more realistic issue here, it
seems to me, and I am reminded of an earlier working group where one of
our consultants said it's the water, stupid, the more realistic thing
that is likely to happen here is that the initial conditions that are
the basis for the TSPA may not be appropriately represented.
MR. BUDNITZ: That's a fair comment.
MR. GARRICK: Because the thing that really distinguishes
this from the reactor case is the fact that the peak dose may not occur
for 300,000 - 400,000 years.
MR. BUDNITZ: Well, they have a 10,000 year requirement.
MR. GARRICK: I don't care. I don't care. I'm a risk
analyst. I am not a regulator, and so the thing that drives that --
there is almost as much of a singularity in the waste disposal problem
as core damage is in the reactor problem in terms of the release, and so
I think that what is really where we are going to find the most
opportunity for having miscalled this thing is not so much with the
design of the barrier but with the initial conditions that are the basis
for the performance assessment in the first place.
MR. BUDNITZ: You could be right.
MS. LUI: Okay. Next slide. There are two technical issues
that we are wrestling with in terms of the multiple barriers analysis.
Basically we mentioned about underperformance of a barrier. What we can
do is we can prescribe what should be the degree of underperformance or
we can take a more performance-based approach. Let DOE look at the
amount of evidence that they have in terms of supporting the barriers'
capability they claim in the TSPA analysis and then they can make a
judgment of what should be the appropriate degree of underperformance
for that particular barrier in the barrier underperformance analysis.
Another issue we are looking at is how should NRC evaluate
the outcome of the underperformance analysis?
MR. APOSTOLAKIS: Which is what I have been saying. You
haven't said anything about the assumptions that the analysis makes. Is
that buried somewhere here?
I don't understand.
MS. LUI: The assumptions for the barriers underperformance
MR. APOSTOLAKIS: Yes, for transport of radionuclides.
MS. LUI: It is all part of the total system performance
MR. APOSTOLAKIS: I understand that.
MS. LUI: Right.
MR. APOSTOLAKIS: But where in this scheme of things do you
worry about the assumptions being wrong?
MR. GARRICK: That's what I mean by the initial conditions.
MR. APOSTOLAKIS: I know, but I don't see where it is.
MR. GREEVES: I think Dr. Garrick would say that that is
included in the original performance assessment. When you step off and
start doing these under performance evaluations, I think you would have
to talk about understanding what those assumptions were and try to
justify why you made those.
MR. APOSTOLAKIS: Right.
MR. GREEVES: The DOE could make a statement this is my
assumption, we think it's reasonable. The Staff could look at it and
say looks good but we have a little wider band. I think that is part of
what we are about.
MR. APOSTOLAKIS: But that brings me back to my earlier
question where I was told that I was impatient. How do you handle model
uncertainty then in the base case? You say known uncertainties are
appropriately captured. What does that mean?
MS. LUI: If part of the consideration of the alternative
conceptual models --
MR. APOSTOLAKIS: But do we know how to do that? Do we
understand the conceptual framework? Do we know how to do that?
MS. LUI: Okay. There are a couple -- there is a stepwise
process. Basically DOE will have to identify what are the alternative,
what are the conceptual models, what are the different conceptual models
that are consistent with all the information that we have up to date and
that they have to make a justification why they have included certain
ones and they have excluded certain ones from their consideration in the
total system performance assessment.
MR. APOSTOLAKIS: What if they take all 11 of them and give
them different weights?
MR. EISENBERG: They can do that, but we would also want to
see that information disaggregated and we would look to see to some
degree what the bounding one would be and we would probably want them to
show compliance with that one.
MR. APOSTOLAKIS: Which each one?
MR. EISENBERG: Yes.
MR. APOSTOLAKIS: With each of the 11?
MR. EISENBERG: No, with whatever the bounding one was.
DR. KRESS: That is each of them.
MR. APOSTOLAKIS: That is each of them, yes, if the bounding
one does it -- it's each of them.
Is that something that people have really thought about?
DR. KRESS: It is not clear to me where you are using
probabilities in this process at all.
MR. APOSTOLAKIS: They are not.
DR. KRESS: That seems to be the shortcoming in this whole
MR. APOSTOLAKIS: That's right.
MS. LUI: Probabilities fall into a total system performance
DR. KRESS: It is part of the performance assessment, I
MS. LUI: Right.
MR. APOSTOLAKIS: Yes, but --
MS. LUI: There are disruptive scenarios that have the
equivalent of initiating events probability and then you have expected
evolution of the repository behavior.
MR. APOSTOLAKIS: We just agreed that maybe in one piece of
this evolution there are questions about the medium, for example, okay,
and we have transport through fissures, fissures or something else, and
I think I heard Dr. Eisenberg say that if there are questions like that
and you have 11 different ways you can go, you better meet the
regulations with each one of them.
I am asking whether this committee has discussed this issue,
because that sounds to me like a license to kill.
MR. GREEVES: I think that there has to be a qualification
on 11. It has to be something that is reasonable. You can come up with
something that is non-physical and that one should be discarded.
MR. APOSTOLAKIS: Well, physical I understand, but how about
MR. BERNERO: You know, I am sorry to hear Norm use the word
"compliance." The total system performance assessment which is supposed
to take due account of uncertainties is being used as a compliance tool,
is the result of it consistent with the objective, the safety isolation
objective as stated?
These are sensitivity analyses and these sensitivity
analyses, somewhat arbitrarily chosen, somewhat arbitrarily applied,
should explore how close to the edge of the cliff of unacceptability
they are or their results would be, and it is not compliance --
MR. EISENBERG: For a particular barrier --
MR. BERNERO: I mean it is license to kill if you say now
change that assumption to the worst case and show me you still comply.
You just made that your compliance case.
MR. EISENBERG: No, I think we are talking about two
different things. I think what George was talking about was how do we
consider conceptual model uncertainty in the performance assessment as a
whole, not how do we do these defense-in-depth calculations.
MR. APOSTOLAKIS: They are related though, Norman. They are
related, very much related.
MR. EISENBERG: I thought how the question was phrased I
thought the predicate for it was that you had 11 different conceptual
models and you had no information to be able to distinguish --
MR. APOSTOLAKIS: Yes.
MR. EISENBERG: -- between one and the other.
MR. APOSTOLAKIS: Well, I didn't say, the second part I
MR. EISENBERG: Well, then do you have a preferred model and
do you have evidence to support the preferred model?
MR. APOSTOLAKIS: I don't know. Maybe there are two or
three possibilities. I don't know. We may do what NUREG 1150 did,
collect a bunch of experts and try to assign weights. I don't know but
I would really question the wisdom of saying that I will do it for each
model and see what --
MR. EISENBERG: But that -- my answer was predicated on the
basis that there was nothing to distinguish between --
MR. APOSTOLAKIS: Okay.
MR. EISENBERG: -- between the different conceptual models.
Now you are telling me you have more information. Well, if you have
more information, you should use it.
MR. APOSTOLAKIS: But is it being used now?
MR. EISENBERG: Yes.
MR. APOSTOLAKIS: Yes?
MR. GREEVES: Both the Staff and DOE have done these
calculations and we have briefed the committee on them.
MR. BUDNITZ: But I am still stuck with, sorry, with my
Let's suppose that we have a barrier and we have enough of a
quantification of our state of knowledge of its performance so that we
can say its performance is in a certain range -- just to be numerical
about it, without knowing quite what it means, it is between 4 and 400,
this is a completely arbitrary discussion, and 400 is worse than 4,
right, and let's suppose we knew nothing more than that. It was a
complete maximum entropy. We said we knew damn well it couldn't be
lower than 4 or greater than 400.
You would be saying, gee, you better assume 400 and show us
it works. I am not disagreeing with that, but if you have a state of
knowledge that says, well, I am sure that it is between 4 and 400, but I
actually have knowledge that tells me that there is a curve and
distribution and the probability it's at either end is really quite
small although it is possible, and we know it is bounded. It can't be
more than 400. Then it is not right -- by the way, if you use 400 and
you still pass, great. You do that every day of the week in every
analysis we know. That is the best way to show it, but it is not right
to insist that when, and I know you understand this, but now we come to
this question about underperformance and compensation.
Are you going to ask for that barrier -- now this is just
very conceptual -- that DOE decide which underperformance number to pick
and then they are going to come and bring you the rock, and the thing I
said, "Wrong rock" or are you going to tell them in advance what your
decision criterion can be so that they can spend more money on a better
design or spend more money on more analysis or something so that they
know going in what they can expect from you, because I think unless they
know that, this process is unsatisfactory for me as a citizen, and I
hope it ought to be unsatisfactory for the Commissioners as the
statutory authority because the Department needs to know the rules and
the speed limit before they submit the application.
MR. EISENBERG: The Department doesn't have its design
finalized yet and it doesn't have its safety strategy finalized yet, so
it can't tell us how much reliance it is placing on different components
of the system.
MR. BUDNITZ: I understand what you are saying.
MR. EISENBERG: I am too. We are understanding each other.
MR. BUDNITZ: It's iterative but those guys have to do --
they are the Applicant.
MR. GREEVES: And those guys did a viability assessment.
MR. BUDNITZ: Yes, I know it.
MR. GREEVES: So they are not without ability.
MR. BUDNITZ: We all know that. We all know that.
MR. BERNERO: But I have got to quarrel with you, Bob, on
the regulator can't take the burden of sharp prescription of what does
it take to prove safety. You can't do that. It is, like it or not, it
is a show me the rock. DOE has the primary responsibility and there has
to be some kind of guidance on what size rocks and what texture.
MR. BUDNITZ: The boundaries.
MR. BERNERO: But at the same time you can't get away from
the fact that DOE has far more capability and far more responsibility to
develop these arguments to show that there is not undue reliance --
MR. BUDNITZ: Bob, I agree with you absolutely, completely
about whose responsibility is where. What I was worried about was that
the amount of underperformance the Department will assume may be way
short of what you would have done and then they have got their design
they have frozen. They are in the licensing process and they could have
fixed it earlier.
MR. GARRICK: Bob, I suspect that if you calculated the
matrix I showed you this morning, the more detailed one, the answer
would be obvious.
MR. BUDNITZ: You may be right.
MR. GARRICK: Yes. If you have the performance of the
individual barriers with and without in context, that to me would be the
strongest piece of evidence you could possibly have for me to make a
judgment about the performance and I know you said in your talk that you
can't remove the barrier --
MR. BUDNITZ: Completely, of course.
MR. GARRICK: -- completely, but you can do variations on it
and, as a matter of fact, as you decompose it into more and more
detailed barriers you can increasingly remove it more easily.
MR. BUDNITZ: That's fair.
MR. GARRICK: And with increasing accuracy.
MR. BUDNITZ: Just as your microscope goes --
MR. LEVENSON: John, as I have been listening to this, I'm
thinking what would bother me about it if I were trying to conform and
this word "compensate" is a very loose end, that it would change
completely what needed to be done if you said adequately compensate as
opposed to totally compensate, and without a modifier there is an
implication of total.
I would give an example. In your base case maybe the dose
to the public is -- I will use Bob's one one-hundredth of what is
allowable, but you fail one barrier and now you are only one-tenth of
what is allowable. Clearly you are way under what is allowable but you
haven't fully compensated and so I think the choice of the word
"compensate" without a modifier is likely to cause all kinds of
MS. LUI: Yes, we agree with you basically. That is why
these are two key technical issues that the Staff is struggling with, to
make sure that the rule and the guidance is going to follow and be
consistent with the Commission's mandate on a risk-informed, performance
based regulatory approach and at the same time provide sufficient model
to the Department so that they will be able to submit a quality license
I think we have kind of skipped over some of the points that
are discussed on the next slide, so proceed to Summary.
MR. GARRICK: Which number are you on, just for clarity's
MS. LUI: Slide Number 8. The multiple barrier requirements
go to be a system requirement.
We shied away from the subsystem -- qualitative subsystem
performance objective in Part 63, in the proposed Part 63 and we will
continue the track that we will keep the multiple barrier requirement as
a system requirement.
In other words, we will not set performance goals for
barriers such as waste package and natural settings.
In our evaluation of DOE's license application, the goal is
to look for that Both the engineered and geologic systems contribute to
safety. That goes back to safety that is not wholly dependent on a
single barrier concept.
I think we have pretty much beaten the second check-mark
here to death --
MS. LUI: -- and the last one is we not seeking complete
redundancy for the barriers.
The last remark is just to reiterate that the public comment
period is over and we are well underway in terms of analyzing the public
comments and providing and preparing the response, and whatever
information that we hear during these particular meetings that will be
available to us in terms of finishing up the final rule and drafting the
Yucca Mountain Review Plan. We intend to put the transcript of this
meeting on the website so that it will be available to the general
MR. GARRICK: Let me postulate a situation. We have learned
a lot from the TSPA work. We have learned so much that where we used to
use the word frequently "geological isolation" we are using it less and
less, because we have pretty much learned that if we have a source term
and it is mobilized, it just delays the transport of that material into
the biosphere. It doesn't isolate it from the biosphere.
At least we haven't been able to characterize, we don't
think we are able to characterize any site where we could achieve
complete isolation in the absence of assistance from engineered systems.
Now supposing somebody came along and suppose they convinced
you that I have designed the one million year waste package and my
confidence in that containment capability is far greater than my
confidence in the containment and transport capability of the natural
setting. Obviously if you have a defense-in-depth philosophy like you
are stating here and that we are seeking balance, which I in principle
kind of agree to, you'd deny them the license.
MR. GREEVES: Why would you deny them the license? You lost
MR. GARRICK: Well, what I am saying, if somebody comes
along with the perfect, with a million year waste package, and there's
engineers that believe they can do that, and yet the geologic setting
they couldn't convince you that if there was a source term that there
would be adequate containment, but with the waste package of course
there is adequate containment, so you don't have the defense-in-depth
but you have a waste package that convincingly will last a million
With Part 6 could you license that?
MR. GREEVES: I think you have carried us too far of a
MR. GARRICK: Well, I don't think it is so far a stretch.
Frankly, I think it is probably much easier to design a million year
waste package than it would be to characterize Yucca Mountain down to
the few meters.
MR. GREEVES: Your dialogue was saying that the site gives
you nothing is the way you --
MR. GARRICK: Eventually it doesn't give you anything. It
gives you dilution. It gives you something.
MR. GREEVES: I don't agree with that statement.
MR. GARRICK: But the one thing that the Nevadans are coming
to us very strong on is, and the NRC is agreeing with them, at least in
the public media, that we are now talking about delay, not isolation.
MR. GREEVES: Anybody that's been in this business, Bob
Bernero said it earlier, it's just a question of time whether it is high
level waste, low level waste. You cannot guarantee containment. There
will be some time when you have to --
MR. GARRICK: The argument being, John, that there's a lot
of people that believe I can do a much better job at building something
to a specification than I can at characterizing a mountain into a level
of detail necessary to give me the same output.
MR. GREEVES: I am aware there are people out there like
that. We are also aware that there is a piece of legislation that calls
for multiple barriers.
MR. GARRICK: That's all I am getting at. That's back to my
MR. GREEVES: The simplest -- an engineered barrier and the
MR. GARRICK: Are we ending up with a law, with a regulation
here where we couldn't license a repository that has overwhelming
evidence that it will retain its integrity for a million years?
MR. EISENBERG: Dr. Garrick, there is no intent to put a
roof on the quality of any barrier. DOE should make each barrier as
good as they can.
MR. GARRICK: That isn't my point. My point is -- MR.
EISENBERG: Well, it sounds like it is your point.
MR. BERNERO: I would like to interject on behalf of the
Staff, as if I was still there.
What you describe is a very good description of the Swedish
MR. GARRICK: Yes.
MR. BERNERO: Which is the sole purpose of the repository
isolation is to maintain reducing chemical conditions so that this very
nicely designed million year package will live for a million years.
MR. GARRICK: Right.
MR. BERNERO: And besides that, that water down there is
fossil water It isn't going to move for a long, long time, and it is a
They of course are a piece of granite that is rising up out
of the sea and you have a choice of granite, granite and granite for a
MR. BERNERO: The United States has a system of laws which
gives us a statutory requirement that says you must have multiple
barriers. It also has a statutory requirement that DOE cannot look at
Now that is not a technically based requirement. It's an
entirely politically based requirement.
There is a system of laws and there is a distinction that
one has to make in what would constitute an acceptable repository as
against what would constitute a preferable or ideal repository. At one
time we had three sites to be simultaneously characterized, and we used
to call it "The Beauty Contest." Insanely expensive. Just imagine
doing Yucca Mountain in triplicate and trying to keep them on the same
What we have to have in the United States is what is an
acceptable repository. It's been accomplished in the WIPP case, warts
and all, you know, and certainly we can talk for hours and hours on what
should have been done there, but it's been done and I am convinced it is
an acceptable repository and warts and all this Yucca Mountain thing --
MR. APOSTOLAKIS: I think it also comes back to the issue of
prevention versus mitigation. Maybe -- I really don't like, to
generalize a little bit, regulatory documents that talk in terms of
number of barriers. In fact, if this subcommittee writes a letter, that
would be a good thing to attack, because it is such a fuzzy concept that
can be misused and so on. I don't know what it means, multiple
barriers, to begin with, and I think a lot of the debates we are having
here come from the fact that the Staff naturally feels that they have to
comply with what the Commission says and the Commission says multiple
barriers, the legislation, I'm sorry. But this is an independent
advisory committee so we can write --
DR. KRESS: Did the Senate say how many barriers was
MR. BERNERO: No.
MR. APOSTOLAKIS: Well, the more I think about it, it's
really the root cause of a lot of emotional debates, because I am not
even sure -- you gave us a good example with the reactor vessel.
Up until this morning I would call it one barrier. Now you
tell it is not one barrier. Now I have no basis of saying it's not or
it is or it is not. I think it's wrong to count barriers, to count
something you have not defined.
MR. LEVENSON: But John, in response to your question, I
think the answer is it could be licensed because the legislation, as I
understand it, does not say that each barrier has to be 100 percent
The legislation just says there must be more than one
MR. APOSTOLAKIS: Which defeats the whole idea, of course.
DR. KRESS: I think at this point -- are you finished?
MR. GREEVES: Let me just summarize. We are finished.
MR. APOSTOLAKIS: Good.
MR. GREEVES: You think I should stop there? He said we
were finished. He didn't say we've had it. I think we have worn it
Just to summarize, I think Norm did a good job of showing
you the spectrum of issues that face us across the licensees that NMSS
has. It is a difficult issue and I think we have learned something from
watching the process here, and I think some things are going to come out
in the future that will help us, and each one of those -- it is almost
like the chart that Norm showed. For each one of those arenas, we have
got to start making some decisions.
You spoke at length about the DOE issue, but each of those
we have got to sort of make some decisions. I know you all appreciate
that the Staff needs to be consistent with the Commission policy and the
legislation, so that is something that we will be holding in our minds
as we draft the regulations.
Something that has come out to me is listening to us all
talk around the room is transparency. I think we have got to find a way
to explain these things that is a bit more clear. I think we talked
past each other on occasion, so I challenge us to -- over time we are
going to have to make this more transparent to other stakeholders.
I do ask you to keep in mind what the Staff presented are
preliminary considerations. We are working under the requirements for
developing the rule process, and I know Bob is disappointed he didn't
see the number he was looking for, but that is something we are about.
MR. BUDNITZ: Doesn't have to be a number.
MR. GREEVES: Well, I think you raised some good points and
I agree with the need to do it one way or the other, and we didn't tell
MR. BUDNITZ: That's fine.
MR. GREEVES: And so those will be my closing remarks and I
assure you we are still considering these issues and we are going to
look at this transcript and I think it will be helpful. Thank you.
DR. KRESS: Thank you very much. At this point I'll take
another break for about fifteen minutes, and that would be be back at
ten minutes, by this clock, after 3:00.
DR. KRESS: We are at the point on the agenda where we are
going to hear from Gary Holahan and Tom King. Our pleasure, gentlemen.
MR. HOLAHAN: Good afternoon. This is Gary Holahan. I am
the Director of the Division of Systems Safety and Analysis in the
Office of Nuclear Reactor Regulation, and Tom King and I are going to
discuss what defense-in-depth means to the reactor program. I think you
will hear a lot of things that you heard this morning, because I think
we are all playing from the same historical book, so some of what we
discuss will be historical, some of it is recent and ongoing activities,
and some of it is looking to the future, so I will start out with a bit
of the historical perspective and Tom is going to cover the future.
I think it is interesting the first point we are making is
that in fact there is no formal regulation or agency policy statement on
defense-in-depth and I think this goes back and is consistent with Tom
Murley's comments this morning about defense-in-depth isn't a rule or a
specific requirement, which I think leaves a little bit to a number of
comments this morning about are we talking about a philosophy or a
policy or a guidance or a rule or a requirement or a commandment?
I guess at that point I would have to agree with Dr. Budnitz
that what really matters is how you implement it, so in fact we have
called defense-in-depth a philosophy, not a specific regulatory
requirement, and in our recent guidance documents we have said that it
is one of our principles that we preserve that philosophy, so George
might be offended. We used the word principle and philosophy in the
same sentence, but luckily George and his subcommittee concurred in that
document, so we'll feel comfortable about it.
MR. HOLAHAN: But it was two or more years ago.
MR. APOSTOLAKIS: Nothing less is expected of Gary.
MR. HOLAHAN: The second point in fact is that as with the
materials program, the reactor program is really working with the same
philosophical concept of defense-in-depth. In fact, we are quoting the
same version that Bob Bernero mentioned this morning where
defense-in-depth, as was said earlier, has successive compensatory
measures and it has this element of not being wholly dependent upon any
single element of the design.
There have been previous definitions of defense-in-depth and
they have all been more or less consistent. I am going to show you a
couple of historical examples in just a minute.
The third point I would like to make on this introductory
slide is that what really counts is that this philosophy, the same
philosophy can be implemented in a number of different ways and what you
see in the reactor program is not necessarily the same thing as you see
in the materials program and I think the agency feels reasonably
comfortable calling both of those defense-in-depth philosophy.
In the reactor program I am going to discuss the regulations
themselves where defense-in-depth is included in the regulations even
though it isn't a specific regulation itself, also how the licensing
process and the license amendment process have dealt with the subject
and the new reactor oversight process, where oversight includes
inspection, enforcement, monitoring of licensee performance, where the
elements of defense-in-depth are embedded in that process as well. Next
Well, you can see on this viewgraph Part 50 includes
defense-in-depth in a number of ways. The concepts of prevention,
mitigation, single failure, redundancy, diversity -- these are all
elements of defense-in-depth. When we talk about it, you can talk about
defense-in-depth in a number of ways. You can talk about physical
barriers. You can talk about functional barriers. You can talk about I
think Tom Kress has suggested a number of times risk allocation in fact
is a defense-in-depth concept. You can put numerical goals on things
like core damage frequency and large early release, and that in effect
is a way of providing defense-in-depth. Next viewgraph.
There are two viewgraphs that are used as part of a training
program that NRC has. It's called "Perspectives on Reactor Safety" and
it is sort of, in part it is a history book that Denny Ross and a number
of people worked on with Sandia to put together so that NRC's new Staff
members have an appreciation of not only what the requirements are but
how they got that way, and it covers sort of the history of the '60s and
'70s as the requirements were built.
As part of that, in fact there is a section on the concept
of defense-in-depth, what it means and how it was developed and I am
going to show you two viewgraphs from that material.
What you see here is one concept of defense-in-depth, which
I think I would call the functional definition. That is, you look at
prevention, mitigation in terms of having safety systems and
containment, and siting and emergency planning. In this particular
example you will see that accident management is also identified as a
level of defense-in-depth. Some people would push it a little bit into
a containment performance issue. Some people would talk about it as an
emergency response issue, but you see how the measures of
defense-in-depth basically show that public safety is protected by a
series of functional type barriers. Tom, can I see the other one?
I think especially years ago people generally talked about
defense-in-depth in terms of physical barriers, and in fact in the
training book these are two pages right together, and so these concepts
sort of grew up together over the years and the concepts of physical
barriers including the fuel pellet and the cladding, reactor coolant
system, containment, and then things like exclusion areas --these are
the physical barriers.
Now what we know is this is a defense-in-depth concept.
Each of these defense-in-depth concepts really has its own sort of
strengths and weaknesses. If physical barriers were the only
defense-in-depth concept, I think we would have come quickly to the
realization that common cause failures and interdependencies make this
an incomplete concept for defense-in-depth. In fact, the functional
concept in my mind is more complete and in a number of ways, using PRA
and whether you call it allocation or other ways of looking at core
damage frequency, even the concept of Level 1, 2 and 3 in PRA in my mind
are a form of defense-in-depth and probably a more complete form.
One of the ways in which the regulations call for
defense-in-depth, and this is just one example that I have picked out,
you could probably find dozens, if not hundreds, of places where a
concept is embedded in the regulations, right in the general design
In fact, it is broken up into six sections. One of the
sections itself is called "Protection by Multiple Barriers" but in
addition to that, the other sections of the general design criteria,
which really play a strong role in determining what an acceptable
reactor design looks like, in fact call for a reactor core that behaves
well, a primary coolant system with low failure probability, and then
fluid systems, either normal ones or emergency ones, to handle failures
and the reactor containment and fuel and radioactivity control really
talks about fuel in the sense of fuel handling, and that doesn't mean
that when it is in the core, it means when it is a potential source, so
the very structure of the regulations down to the general design
criteria have embedded in them a defense-in-depth concept.
I think I said I would talk about licensing but I think I
skipped -- let me do the oversight program and then I'll talk about the
license amendment process because that is one that we have been changing
lately and it has a good kick-off point for Tom to get into our more
The reactor oversight process was really given almost a 100
percent overhaul in the last year, where the inspection program, the
enforcement program have basically been totally rewritten, and they have
been rewritten with two concepts in mind. One is to be more
performance-based, to look at licensee performance and react to it, and
the other is to use more risk insights in the process, but in doing so
the defense-in-depth concept is being preserved by the use of what are
called cornerstones, and I am going to show you how the cornerstones fit
into the process.
Basically the message is that the cornerstones in the
oversight process are the ways of embedding defense-in-depth.
Cornerstones are defense-in-depth features and in fact if you read the
papers on the subject, the concept of defense-in-depth comes up in a
number of points.
This is a viewgraph that many of you may have seen before.
It is used in a lot of the presentations on the oversight process and if
I can lead you from the top down, public health and safety really means
that we worry about how the reactors behave and radiation safety, both
in terms of the public and workers. That's the Part 20, Part 100 type
issue, and safeguards, so the issues to the right are really in addition
to what we have talked about most of the day in terms of public health
and safety from unusual type of severe accidents.
If you will look at the way the program is structured,
reactor safety has four basic elements to it. They are called
cornerstones but you could have called them defense-in-depth elements if
you wanted to.
We look at initiating events, mitigating system performance,
barrier integrity, and emergency preparedness, and those are basically I
think a combination of functional and physical barriers.
The way the oversight process works, the licensee
performance, both in terms of performance indicators and inspection
results from our inspections staff are put into these categories, and
then we make judgments about the licensee performance in those areas.
If you go to the next slide, I can continue.
I am going through this kind of quickly, just not to explain
the whole process to you but just to show how the concepts,
defense-in-depth concepts, are built in here.
The performance indicators as used in the reactor oversight
process are in fact groups together depending on which of the
cornerstones they relate to, so things like reactor scrams or
significant initiating events and transients, they go into the
initiating event cornerstone, and things like the safety system
performance and unavailability, those go into the mitigation system, and
so the licensee performance in terms of performance indicators and
inspection findings are measured with respect to thresholds to identify
their significance and they are folded into these cornerstones. We can
go to the next one.
In fact, I am not going to discuss this viewgraph. Just for
completeness it shows how each of the cornerstones has indicator input
The next viewgraph is a little hard to follow, but the basic
concept is across the top you will see a spectrum of results in which
various levels of performance of increasing safety significance are
monitored, and so on the extreme left what you will see is everything is
pretty normal, and that is the inputs to the cornerstones, each of the
cornerstones, not just public health and safety sort of dose limit, but
each of the cornerstones is performing well.
If you will look down that column it says we have a routine
inspection program and licensee fixes issues on their own, and sort of
everything runs sort of normally and this is, you know, we use the
terminology of "green" -- this is normal green performance in terms of
for a licensee. As you move to the right, across the top columns, you
will see increasing level of concern, and that is indicated by degraded
performance in one or more cornerstones.
As you can see, as it sort of escalates, it is not
only that the total licensee performance seems to be unacceptable in
some way, but the NRC response will escalate when the performance in one
cornerstone area becomes of increasing concern to the level of being
warranting interactions at Regional Branch Chief level, Regional
Division Director, Regional Administrator, EDO and even getting to the
point of the Commission.
So what it says, and there's lot of detail on here that I am
not going to cover today, the basic message is we are looking at
licensee performance at the cornerstone, but that's basically at the
defense-in-depth functional levels and making judgments about how well
the licensees are doing, what level of interaction we ought to take with
them, whether their performance looks normal and we ought to sort of be
restrained and allow them to deal with their own issues, take corrective
action when problems occur, or whether a higher level of management
involvement and more extreme expectations are appropriate
Now the system is set up basically as an early warning
system. It is not so easy to go from green to red. Part of the
workings of the systems is you expect the licensees to know very well
what the rules of the game are. If their performance begins to degrade,
they know it early-on. We expect them to be dealing with it early. We
don't expect licensees to be in the yellow and red area because there's
plenty of warning for them to turn things around, but the scheme shows
how the Staff will be responsive to cornerstones or defense-in-depth
weakenings, and in fact potential failures. Tom?
I know that is kind of a lot to digest. The only point I
wanted to get across is that even though defense-in-depth is not written
as a regulatory requirement it has a value as a guiding philosophy and
it can be built into various programs in a practical and usable manner.
Now in the license amendment process we have developed
Regulatory Guide 1.174. Even though 1.174 has a lot of general safety
philosophy in it, it was really meant as a licensing amendment guidance
document and there are five safety principles associated with deciding
whether a license amendment change is acceptable or not.
I know the ACRS members are very familiar with that. We
spent a lot of time with the committee on these issues and if my memory
is accurate, and I think it is, even the concept of having five
relatively high level safety principles was a concept that came up at
this table in the interactions between the Staff and George, your ACRS
One of those five principles is that there ought to be a
defense-in-depth philosophy and my recollection is we talked a long time
about this issue of should there be defense-in-depth, should there be
defense-in-depth philosophy where we are talking about never giving up
any measure of defense-in-depth, and I think it was an important issue.
I think in a number of ways it still is an important issue and I think
next month we will talk about ACRS has a session on impediments to
risk-informed regulation, and I know a lot of people are concerned that
this is a potential impediment, and I think we have certainly got it on
our list of one of the things we want to talk about.
Reg Guide 1.174, its corresponding Standard Review Plan, and
the related documents on how to do risk-informed regulation not only
mention that there should be a defense-in-depth philosophy but give you
some insights as to what that means and it identifies issues like
balance between prevention and mitigation, avoidance of over-reliance.
Now these are general concepts. They are not numerical values. I think
George has expressed the idea that you should be very careful about not
counting the numbers of defense-in-depth or try to quantify it too much,
and I think we recognize the danger in doing these things.
Those concepts are discussed in the guidance documents. I
think it clearly says we are not trying to assure that there is no
change in the level of defense-in-depth. What we are saying is there
should be no change in the philosophy. So if a licensee wants a license
amendment to remove the containment, they ought not to bother because we
are not going to pursue that.
MR. APOSTOLAKIS: One important point here, which I believe
is an assumption on your part and most people when they talk about these
things is you are talking about these issues for the current generation
of nuclear power plants.
There is a certain assumption here that -- in other words,
would you be as absolute in rejecting a request for no containment for
any future reactor? I doubt that, because you don't know what physical
pieces of those --
MR. HOLAHAN: I wouldn't reject it categorically.
MR. APOSTOLAKIS: So this is really for the current
generation, which is I think a reasonable thing to do.
MR. HOLAHAN: Well, for the current generation and I think
for the evolutionary and advanced reactors that we have seen.
MR. APOSTOLAKIS: Yes. I agree.
MR. HOLAHAN: But I think this ought to be left as a
relatively high hurdle.
MR. APOSTOLAKIS: I agree.
MR. HOLAHAN: Okay. By its nature, what we are trying to do
in the reactor area, and I recognize that in the materials area there
are some other considerations, we are providing a very high level of
protection, that is very low probabilities for high consequence events.
Almost by definition, if that is the arena that you are in, you are not
going to have a lot of experience to deal with and you are going to be
extrapolating from pieces of what you know, and issues like completeness
and modelling are going to be difficult ones.
One of the things that I sort of keep an eye on is the
accident sequence precursor program, previously in AEOD, now in the
Office of Research, and my recollection of if not the last but one of
the recent Commission papers on that program, maybe a year ago or so, I
think it said something like half of the accident sequence precursors,
the ones of some significance, were things that were not previously
modelled, and so the signal is we are still at a time in which there are
surprises to be had, and by its very nature, you know, you are going to
have to develop an awful lot of operating experience before you get to
the point in which you say my modelling and my completeness are minor
MR. APOSTOLAKIS: Well, again, I would put some qualifiers
to what you just said. What does it mean it's not modelled? I mean
maybe the exact sequence of events was not modelled but maybe it is a
subset of something bigger that was modelled.
MR. HOLAHAN: Well, I think --
MR. APOSTOLAKIS: I agree with that.
MR. HOLAHAN: I think it is worse than that.
MR. APOSTOLAKIS: I think in some instances it might be, but
the other, I mean in all fairness you should also mention then the very
important findings of the former AEOD people that the system
unavailabilities they find are either -- are within the range of values
of PRAs found --
MR. HOLAHAN: Yes.
MR. APOSTOLAKIS: -- which is really a very good
confirmatory piece of evidence that what we are doing is not off the
MR. HOLAHAN: And in general initiating event frequencies
are somewhat better and in fact in my mind, more important than either
of those is that common cause failures are lower than is generally
MR. APOSTOLAKIS: Right. They are going down and they are
MR. BUDNITZ: You are looking under the lamppost some of the
time because half of the risk overall of the fleet comes from fires and
earthquakes and configuration compromises that would make you more
vulnerable to fires and earthquakes are not modelled in ASP today, as
George and I know, since we wrote a NUREG about it which hasn't been
MR. KING: But there haven't been that many fires and we are
MR. BUDNITZ: Well, there haven't been fires or earthquakes,
but we are talking about configuration compromises that will make you
more vulnerable if you had one.
MR. KING: Yes.
MR. BUDNITZ: Those happen all the time.
MR. KING: There haven't been any earthquakes.
MR. HOLAHAN: And my recollection is isn't that issue number
one of twelve that we are dealing with in the risk-informed fire
MR. BUDNITZ: I hope so.
MR. HOLAHAN: I think it is on top of the list.
MR. BUDNITZ: I hope so.
MR. HOLAHAN: So the message I want to leave you with is in
the reactor area for the plants we are currently dealing with, which
basically are operating plants -- not so long ago we dealt with advanced
reactor designs -- but in this context I don't think they were all that
Defense-in-depth has been an integral part of our decision
process, what we envision for risk-informing Part 50, and Tom is going
talk to Option 3, but certainly if I remember the ways the options are
set up for risk-informing Part 50.
Option 1 is just to continue with some of the rulemakings
that we have ongoing, 50.59 and maintenance rule and things like that.
Option 2 is to take those issues related to day to day
operational performance and parts of the plant that get special
treatment in operations, things like quality assurance and technical
specifications, and maintenance type activities, and to risk inform
those sort of operational type activities.
In doing so, we intend to preserve the current design basis
and that means that the level of defense-in-depth in the plant probably
is not going to be changed very much, and also the other important
characteristic is in deciding what is of safety significance, because in
effect what Option 2 is going to do, it's going to take the old model of
safety-related and not safety-related, something that John Garrick
mentioned this morning, that the PRA world, the risk analysts don't care
much about, and it's going to look at what is risk-significant and what
is not risk-significant.
It's going to overlay those two concepts but in deciding
what is risk-significant or not, we are going to use a concept somewhat
akin to the maintenance rule expert panels where not only are we going
to use the risk analysis numbers, whether it's bottom line numbers or
importance measures, we will use the insights from experienced plant
people who can bring some defense-in-depth and safety margin thoughts
into that process, and we are developing some guidance as to what sort
of things they ought to be thinking about in doing that.
So my message is we currently have defense-in-depth in the
reactor designs, it is in our programs, it is even in our, what I would
say is our most modern risk-informing programs have the concept of
Tom is going to talk about Option 3.
If I look about where we are going with risk-informed
license amendments and those sort of changes, there is a challenge on
the table for us.
I don't think we are going to quantify how much
defense-in-depth you need but we may put some more guidance in place as
to how to deal with issues where maybe it looks like we are either doing
-- I mean I must say I haven't heard any "too littles" but maybe we are
doing too much to preserve more defense-in-depth than a more
risk-informed insight would tell us is necessary.
So the program is ongoing. Defense-in-depth is a -- call it
a philosophy or a guidance concept, and it's basically built into where
MR. APOSTOLAKIS: But the point though, Gary, is that it is
not whether one should have that philosophy and whether one should
ignore, for example, the items you have under Regulatory Guide 1.174.
The question is not what role the risk -- the PRA methods we
have should play here, and I would say, for example, if I took -- given
the evidence that I have including the AEOD evidence, that PRAs have
done a pretty good job modelling system unavailability for individual
safety systems, there is strong evidence that we have done a hell of a
job, then again from my point of view that means that maybe the issue of
unquantified uncertainty is not that important there, although you might
make the point that under severe accident conditions we haven't seen
those and so on but let's take that -- so I would say that now I have a
good tool in my hands to take the seven or eight items you have there
and optimize my operations, optimize my design, and I don't really have
to have a diverse train for example because I manage to achieve the
required levels or the inspected levels simply with redundant trains.
I can make a good case that I have handled common cause
failures and so on, so I suppose the heart of the matter here is is
there anything that will stop me from doing that, another input, another
principle, a philosophy that will say, yeah, you can do all these things
but boy, I really want all seven, and what I am saying is I am not
willing to drop all seven, but first of all if you try to drop them you
will never achieve the numbers you want.
MR. HOLAHAN: Yes, that's right.
MR. APOSTOLAKIS: And second, all I am saying is these are
guidelines. It is a philosophy that you would like to have at your
disposal and use it, but now you have this tool which is reliable in
this particular context, so, you know, I can afford maybe to drop one or
I can afford to minimize the role in one place versus another and so on
and I think that is really what we are doing with the case specific
MR. HOLAHAN: Yes.
MR. APOSTOLAKIS: So this is a good example in fact of a
case where the PRA, it's almost risk-based here, where risk is the
MR. HOLAHAN: Well, I think what I would say is if you go
back and read the section on defense-in-depth in 1.174, I think it's
okay, but that does not mean that in implementing it we won't run into
some tough cases, okay?
MR. APOSTOLAKIS: Sure.
MR. HOLAHAN: And we may be better off just fighting over
those cases than trying to write a guidance document that avoids any
fights in the future.
It may not be possible to write the definitive set of
guidelines on defense-in-depth that never has a problem.
MR. APOSTOLAKIS: And I realize that but I think some sort
of a high level discussion of these issues probably would be beneficial
because I agree that we can't really be too specific at this point.
MR. KING: Reg Guide 1.174, if you recall, in the
defense-in-depth discussion does talk about using PRA, not to do away
with defense-in-depth but to optimize how you achieve it and in effect
in Option 2 and Option 3 risk informing Part 50 it is the same
philosophy, the same approach we are taking.
What I was going to talk about is Option 3 and what are we
doing in our technical study or study of the technical requirements, how
are we folding in defense-in-depth considerations and melding them with
PRA considerations, because for all the risk-informed activities what we
are talking about is not a risk-based approach but using PRA to
complement our traditional way of doing business, which includes
deterministic analysis and defense-in-depth considerations, so we are
trying to keep that approach in both Option 2 and 3, and I will talk to
you about what our thinking is today for doing that under Option 3.
The last piece of this viewgraph I am not going to talk
about. You are going to get a separate presentation on that at some
point in the next month or two from Joe Murphy, but again the reactor
safety goal policy discusses defense-in-depth and we had identified that
as an item for consideration for modifying the safety goal policy.
Perhaps it needs to be updated, expanded, and so forth, consistent with
the risk-informed regulation thought process that we have gone through
in discussion there.
Maybe I'll just take one more minute for background,
particularly for the folks from ACNW on what is Option 3, what are we
trying to do. As Gary mentioned, NRR is working on a rulemaking now
that's called Option 2 that is basically looking at the scope of what
ought to be regulated based upon risk insights and that is in the sense
of special treatment rules -- by special treatment, what should get QA,
what should get equipment qualification and so forth.
The functions would have to remain the same but maybe
depending upon the risk associated with -- the risk significance of the
various systems, structures and components, maybe they don't need the
pedigree that they are receiving today, but again the functions would
all have to be accomplished.
What we are doing under Option 3 is going in and looking at
the functions, the design requirements, what changes should be made
there based upon risk insights.
Maybe to put in context what you are going to hear, the
Option 3 study is going to take place during this calendar year,
calendar year 2000. We are in the initial stages of getting started.
What you are going to hear about is work in progress today. Some of the
details have to be worked out.
What you are going to hear about today we are also going to
put out for public comment fairly soon and we have a workshop, public
workshop, scheduled the end of February to talk about this as well as
the other things we have been working on in the Option 3 study, so this
is subject to a lot of comment and a lot of further discussion. This is
not cast in concrete at this point.
In trying to do the Option 3 study we did realize we had to
come up with what we call a working definition of defense-in-depth,
something that the folks looking at the regulations and the Reg Guides
and the SRPs can take and take the risk insights and sit down and make
some decisions on does what is in there look okay or are some changes
So what we wanted to do was basically develop an approach
under this working definition that would consider defense-in-depth that
traditionally provides some multiple lines of defense -- are not calling
them barriers, we are not counting barriers -- provides some balance
between prevention and mitigation and provides a framework by which we
can address uncertainties in the various accident scenarios, so that is
sort of the scope of what we thought this working definition ought to
There are two elements to the working definition. One,
which is probably the structuralist element, that in our view there
ought to be some floor on defense-in-depth regardless of what your PRA
says, there are probably some things you want to retain, just call it
deterministic or engineering judgment, and then beyond that, there would
be the rationalist piece or implementation elements that can vary
depending on the uncertainty and the risk goals and so forth.
MR. APOSTOLAKIS: This is the pragmatic preliminary proposal
MR. KING: Yes.
MR. APOSTOLAKIS: Structuralist at the high level and
rationalist at lower levels?
MR. HOLAHAN: The rationalist-informed structuralist
MR. KING: It doesn't have to be one way or the other. They
each have some advantages.
MR. APOSTOLAKIS: No, but this is the compromise we came up
with, otherwise the paper would never have been published.
MR. APOSTOLAKIS: Isn't that right, Tom?
MR. KING: Yes.
MR. APOSTOLAKIS: This is the pragmatic.
DR. KRESS: That's pretty much we covered.
MR. APOSTOLAKIS: High level structuralist and --good.
MR. KING: On Slide 15, it talks about the fundamental
pieces or the structuralist pieces. We want to build upon the
cornerstone concept that Gary showed, particularly building upon the
first four cornerstones that are affected by reactor design, initiating
events, prevention and core melt, containment of fission products, and
emergency planning and response.
We feel that this working definition ought to address those
things. We feel that there ought to be some, in the prevention side
there ought to be some again I will call it a floor on design features
that prevent core melt and whether we call those -- we put back in the
single failure criteria or somehow specify some redundancy or diversity,
we haven't worked out exactly the wording of that, but we would not rely
strictly on a risk number to say I have got a highly reliable system,
therefore I don't need any redundancy, diversity, single failure
protection and so forth.
Again, other things you have to consider are how do you
factor the human in and the active versus passive failure, particularly
if we are into the single failure question which in the past has always
been limited to an active component.
We feel that we should retain the ability to contain fission
products given a core melt, that that ought to be a fundamental concept
of part of this working definition and emergency planning and response
ought to be retained. Clearly emergency planning and response is also
affected by siting criteria if you are talking about new plants, but for
existing plants it is pretty well fixed.
Now in addition to assuring the prevention and mitigation we
wanted to assure a balance between the prevention and mitigation and we
felt that we needed to be consistent with the subsidiary risk guidelines
that were developed and used in Reg Guide 1.174.
Those actually came from Commission guidance that we
received over the past years where they gave us a 10 to the minus fourth
core damage frequency damage goal to use and then we developed, as part
of developing Reg Guide 1.174 worked backwards from the safety goal
quantitative health objectives and came back and developed a 10 to the
minus fifth large early release frequency goal that we felt was a good
design objective that if it was met would ensure you would meet the
quantitative health objectives.
MR. GARRICK: In your use of a mitigation here, does it
reach to consequence limiting? In other words, if you are having a goal
with respect to a large early release, now you have material. What do
you mean by mitigation beyond the usual engineered safety features or do
you mean anything beyond that?
Do you include consequence limiting?
MR. KING: The large early release, the word "large" has no
limit on it. It can be a large release --
MR. GARRICK: You are not including --
MR. KING: No. It can lead to early fatalities offsite.
DR. KRESS: Yeah, but it does include emergency response
measures for --
MR. KING: Sure.
DR. KRESS: -- for this LERF to be equivalent to the early
fatalities so that is in there.
MR. KING: Credit is given -- yes -- credit is given for
DR. KRESS: Credit is given for emergency response.
MR. KING: But there is no limit on what large should be.
MR. GARRICK: Well, I am also thinking of fission product
cleanup, retention --
MR. KING: Well, maybe I ought to say a little bit about
large early release. It is not large if it is cleaned up.
DR. KRESS: Yes.
MR. KING: In other words, if it goes through the
suppression pool and scrubbed, it is not considered a large release
because not much gets out of --
DR. KRESS: Those things are inherent in the definition.
MR. GARRICK: Yes, but I am getting at the 10 to the minus
MR. KING: Yes. That is for unscrubbed stuff.
MR. GARRICK: Unscrubbed, yes.
MR. KING: And it can lead to early fatalities.
MR. APOSTOLAKIS: It is directly related to early
MR. HOLAHAN: In effect what happens is if you have a
scrubbed release or a late release or a minor release in fact core
damage frequency 10 to the minus four by default becomes its limit.
DR. KRESS: Yes.
MR. GARRICK: Yes.
MR. KING: Okay. The next thing we have done was say okay,
for this bottom piece what does that mean in terms of looking at the
cornerstones and some practical guidance when you want to go in and look
at the regulations?
We developed sort of a chart that works its way down from
the cornerstone concept and in fact I guess it is a high level
It is not intended to get down to the individual component
or system level. This is to be looked at as fairly high level guidance
but the idea is the following, that you have got various initiating
events and they have various frequencies associated with them.
Some of them are things that you know are going to happen --
loss of offsite power, turbine trips and so forth. They are fairly
frequent and then there is the more infrequent initiators, the large
LOCAs, the large reactivity insertion accidents and so forth, and then
there's the rare events that today aren't included in the design -- the
vessel rupture, the steam generator rupture and so forth.
You can have a list of those, and you can have an estimate
of their frequency and their uncertainty distribution that goes with
And then you want to look at, for each of those, how does
the plant ensure that the core damage frequency and the large early
release frequency is met?
And the idea is that for the more frequent initiators, you
want to be able to have systems in the plant that respond with a high
degree of reliability; so that when those things happen, you're assured
you still meet your 10-4 core damage frequency, and you still have a
robust containment that will meet your LERF goals.
For the things that occur less frequently, maybe you don't
need as much in terms of highly-reliable systems, but the combination of
the two still ought to ensure that you meet your core damage frequency
goal, and you still want to be sure to have containment with the same
degree of protection.
And you still have emergency planning out here, for which
you get some credit.
MR. KRESS: That second line there, does that imply you have
different responses to those initiators, for example, shutting down the
power or the emergency cooling to prevent core damage? You'd have those
If you had to have them for the infrequent initiators, you'd
have to have them for the more frequent ones also. I don't understand
MR. HOLAHAN: It may turn out that way, but, in fact, for
example, you might find that for large loca you need, you know, low
pressure injection, and ECCS accumulators, but for small locas, you only
need the high pressure injection system.
MR. KRESS: I see what you mean.
MR. HOLAHAN: So that says redundancy in high pressure
injection is very important, valuable, but redundancy in those other
systems may not be so important.
MR. APOSTOLAKIS: One comment on this: This would work well
for the so-called internal events.
Now, if you have an earthquake, and possibly a fire, or any
external event that could affect elements of prevention and mitigation,
somehow we need to have maybe a different approach and rethink the
concept of mitigation versus prevention of those big, common-cause
MR. KING: Common-cause failures, yes, how you apply these
to common-cause failures, and how to you apply these to something like
steam generator tube rupture.
MR. APOSTOLAKIS: Although one could apply the same approach
to the sequences that are initiated, perhaps, by the fire, for example,
and have certain requirements in the initiator frequency, the systems
that will mitigate it.
But somehow these two dashed-line boxes come together when
you have those big --
MR. HOLAHAN: I think I agree with you for seismic, but for
fire and flood, I think you can deal with these. In fact, more modern
plants, and certainly evolutionary and advanced plants have dealt with
fire and flood in terms of separation, which allows this to work out
What we see is that fire protection for older plants,
barriers, fire barriers and things like that, are ways of getting
isolation, even though it's not as complete as you see in the modern
With seismic, everything shakes at the same time, and so you
have to deal with that maybe a little differently.
MR. GARRICK: An important part of the large scope PRAs were
the recovery models that were employed. Does the respond include that?
MR. APOSTOLAKIS: Yes. Human recovery actions --
MR. GARRICK: Are over on the right.
MR. APOSTOLAKIS: -- respond to prevent core damage.
MR. GARRICK: Well, also things like recovery of offsite
power, recovery of --
MR. KING: They're in both of these boxes here. And that's
when you go in and look at the --
MR. APOSTOLAKIS: Even prevent initiators. An initiator is
a complete blackout, and human actions to recover diesels and so on is
part of it.
MR. HOLAHAN: I think Dr. Kress made a good point this
morning. Some of these differentiations are a little bit arbitrary. And
whether you say mitigation is mitigation of an initiator, or whether it
is mitigation of core damage, you can break this into finer pieces if
you like, and so a little bit of it is terminology.
MR. KING: The other thing this will help you do is, when
you have something like a steam generator tube rupture where you now
have lost the containment barrier, you've got some frequency associated
with it, and this now becomes one.
That tells you I better have some fairly highly reliable
systems to be able to deal with that.
MR. APOSTOLAKIS: So the message you are sending here, Tom,
is that one cannot really have goals independently of the accident
And what really matters here is really what you have there,
MR. HOLAHAN: And defense-in-depth --
MR. APOSTOLAKIS: And the allocation issue, depending on
reality, on preferences --
MR. HOLAHAN: And defense-in-depth doesn't mean equal
allocation among cornerstones or defense levels. But it means you don't
MR. APOSTOLAKIS: And even there is a seismic issue that
maybe doesn't even allow you to do this, right? So depending on the
Now, why on the performance indicators that the oversight
process uses, sequence or site-specific?
MR. HOLAHAN: Are they are aren't they?
MR. APOSTOLAKIS: Why aren't they?
MR. HOLAHAN: Oh, they are.
MR. APOSTOLAKIS: They are not.
MR. KING: The data is site-specific. The indicators and
the thresholds are generic right now.
MR. APOSTOLAKIS: Yes. The thresholds are generic.
MR. HOLAHAN: The thresholds are generic.
MR. APOSTOLAKIS: Would it be consistent with this approach
to have site-specific thresholds?
MR. HOLAHAN: Well, I think that just -- that would be nice,
but it's complicated. What we've committed to is, in the process where
there are inspection findings or events, we will use as part of this
process, what's called the significance determination process.
MR. APOSTOLAKIS: Yes.
MR. HOLAHAN: And we've committed to that process basically
MR. APOSTOLAKIS: But isn't it true that in the maintenance
rule, the licensees themselves set the goals?
MR. HOLAHAN: Yes.
MR. APOSTOLAKIS: Why can't we ask the licensees to set
goals for their plants for each of the performance indicators? What's
different? Why can't we do it? Somehow we are scared of it.
And then we review it and say fine, or we say change this
and that, and let them do the work. You don't want to do that for 140
MR. HOLAHAN: Well, we did it once.
MR. APOSTOLAKIS: Well, in fact, why don't you build on the
maintenance rule, and say, you know, for a San Onofre, this is what
they're using now for the trains, and San Onofre can --
MR. HOLAHAN: I'm not sure that that level of refinement is
MR. KRESS: I don't think you can justify that level of
MR. APOSTOLAKIS: I think you can.
MR. HOLAHAN: If you think of the scarcity of data, if a
reactor has, you know, four reactor scrams in the same year, whether
it's this type of reactor or that type of reactor, or something, you
know, something funny is going on.
MR. APOSTOLAKIS: I'm willing to grant you that, yes, for
several indicators, probably a generic number would be good enough.
But what I'm questioning is the philosophical approach. I
mean, this is really great.
But when it comes down to actually regulating and
interacting with the licensees, we are switching and going to generic
numbers as a starting point.
MR. KRESS: This thing comes very, very close to what I had
in mind by the allocation process as meaning the defense-in-depth.
Let me ask you a strange questions, Gary: That fourth box
up there, emergency planning and response, with the .1, if that box
wasn't there, and you still had to meet a safety goal that was early
fatalities, your LERF would simply be 10-6 instead of 10-5, I think
, because that .1 is about the mitigation you get.
MR. HOLAHAN: Yes.
MR. KRESS: Do you think all of the plants out there could,
at their present time, meet a LERF of that value?
MR. HOLAHAN: This is a side discussion that Tom and I had
this morning while the discussion was going on. I think it came during
Bob Bernero's presentation.
MR. KRESS: Yes.
MR. HOLAHAN: In general, most of the studies we've seen --
and you've got to recognize that there is completeness and uncertainties
and all those sorts of issues.
Most studies show that current generations of plants meet
the safety goal. That's a little bit of a funny thing to say since we
don't have a safety goal for each plant, but if you extend the concept,
they meet it. And they usually meet it by a factor of more than 10.
So I would think that if you took out a factor of 10 or 20,
which is not unusual, right, for a credit in evacuation, you would be
close. Whether it would exceed the safety goal, maybe not on paper, but
in reality, it would be close enough so that maybe you would say you
couldn't -- you don't really know, right? That's about as close as I
MR. KING: The assumptions that went into NUREG 1150 where
they actually modeled emergency planning, they were based upon looking
at some historical information, chemical spills and so forth, how long
did it take to move people.
And they assumed some lag time from the time the accident
started and you notified people, till they actually moved. And people
moved at a pretty slow rate, and they assumed 95 percent effectiveness
of the evacuation. They didn't assume everybody got out.
And then you see the resulting QHO numbers that came out of
MR. HOLAHAN: And basically, if I remember them correctly,
Tom, and you would know better than I do, my recollection is that if you
moved, you didn't get a lethal dose, right?
I mean, if there were any fatalities, it came from those
left behind, not from some fraction of the people that moved.
MR. BERNERO: I'd like to go back. This is long ago, and
the Sandia siting study in the early 80s had the large early release
PWR-1 or BWR-1 release postulated, and then looked at all the sites that
were proposed or actually selected.
And my recollection is that the site remoteness and
meteorology alone gave you, without -- and I don't remember what the
modeling of emergency response was, if any -- but it gave you .1 for all
sites but Limerick, Indian Point I, and Zion.
MR. APOSTOLAKIS: But wait. I thought the safety goal said
that you postulate the individual is just outside the boundary.
MR. HOLAHAN: No, it's the average.
MR. APOSTOLAKIS: So it doesn't matter how far you are.
MR. BERNERO: What I'm saying is, is there defense-in-depth
that comes from site remoteness?
MR. APOSTOLAKIS: No. The way we're calculating the risk
MR. KRESS: If you had a societal goal.
MR. APOSTOLAKIS: If you had a societal goal --
MR. BERNERO: I'm not talking about goals; I'm talking about
actuality. Right there, there is a box, Emergency Planning and
Response, and it says .1, .1, and that is the defense-in-depth factor or
share that is provided by emergency planning and response.
And what I vaguely recollect is that there was a calculation
that said the site, the remoteness and the meteorology are such that the
typical reactor site provides you .05 or something like that, and only
Limerick was .25 or something.
MR. GARRICK: Well, another study that I recall indicates
something when there was all this debate about the exclusion zone and
what it should be and what was the technical basis for the 10-mile, of
which there wasn't one, some analyses were done, and it turned out that
on a couple of plant-specific cases that some 95 percent of the acute
fatalities occurred within a mile and a half of the site.
MR. HOLAHAN: There is also a quirk in the way that these
are calculated, and I think Dr. Kress, you had an ACRS staff member do
some calculations not so long ago.
And every one of these calculations basically shows that the
value is .06, which means a 1/16th sector around the plant, and it's
driven by a modeling of where does the plume go and who gets affected
and who doesn't.
MR. KRESS: Right.
MR. HOLAHAN: So, it's a little bit of an odd issue.
MR. APOSTOLAKIS: Go ahead. You've given me an idea for
now. I think you should make the last column 1.
MR. KRESS: That was the suggestion that I made this
MR. APOSTOLAKIS: Because you're supposed to postulate that
that individual is at the perimeter of the site. So emergency planning
should have nothing to do with risk calculations.
MR. KRESS: That was the suggestion I made this morning.
MR. HOLAHAN: That's not a PRA.
MR. KRESS: That's a --
MR. APOSTOLAKIS: You're saying, I don't care whether you
MR. HOLAHAN: That's not a PRA.
MR. APOSTOLAKIS: The Commission says, put this guy there,
and tell me what is the probability of death.
So we want it both ways. We don't want to have a societal
health objective, but we want to take advantage of it.
MR. KING: The meteorology still affects that.
MR. HOLAHAN: Those are PRA numbers.
MR. APOSTOLAKIS: But it's the way PRA calculates. PRA
takes the actual population, divides by the number.
MR. KRESS: George is saying we need other risk acceptance
criteria besides the --
MR. APOSTOLAKIS: How can evacuation affect individual risk?
MR. HOLAHAN: It' can't.
MR. APOSTOLAKIS: It can't.
MR. HOLAHAN: You can't evacuate 95 percent.
MR. KRESS: In reality, we do have implied other risk
acceptance criteria, and one of them is involved in that.
MR. APOSTOLAKIS: I think we should rethink this .1, without
MR. HOLAHAN: The problem is that you can't evacuate 95
percent of a person.
MR. APOSTOLAKIS: That's correct.
MR. HOLAHAN: They're either there or they're not.
MR. APOSTOLAKIS: If you read the statement from the
Commission, it very clearly says person within one mile. You can't say
I have an average in one mile.
MR. HOLAHAN: Well, average.
MR. APOSTOLAKIS: The definition of the individual risk is
the probability of death of a postulated individual someplace. But
somehow it has been modified over the years.
MR. BERNERO: It's a one-mile annulus.
MR. HOLAHAN: Yes.
MR. BERNERO: The point I'm concerned about is, if what is
looking for a balance between prevention and mitigation, considering the
cornerstones; that there is a part of the emergency planning and
response cornerstone that comes from just being there in Lower Alloways
Township, New Jersey or wherever the plant is, that even if you said you
don't have to have emergency planning anymore, or we'll just give you a
telephone call and do the best you can, that there is a level of
mitigation that comes from siting remoteness and low population.
MR. HOLAHAN: Yes, I mean, that's true.
MR. BERNERO: And in the future, that could change.
MR. HOLAHAN: Yes. As a matter of fact, my recollection is
the study done by Rick Sherry showed that the safest site in the country
was St. Lucy, and it had nothing to do with the population; it had to do
with it being on the ocean and which way the wind blew.
MR. BUDNITZ: I have two comments about earthquakes, and
they're really very different, and you have to listen to them both.
The first is that, for sure, the very large earthquake --
we're talking about the earthquakes that cause trouble for plants, which
are much bigger than any earthquakes we've even had in California.
They're very large earthquakes, and I hope everybody understands that.
The earthquakes at any site, not just California sites, that
are bigger than the 1906 San Francisco earthquake, that magnitude,
they're very large earthquakes.
And for sure, that last column has got to be one for those
earthquakes. You can't count on evacuation for them, so you have to be
very careful for earthquakes, what you do there, and be sure not to be
The second point, and this is from the PRAs:
If you look at the LERFs from the seismic PRAs -- and I have
probably studied that more than most of the people in this room, and I
plead guilty to that -- they come from two kinds of things:
Part of it comes from very large earthquakes, you know,
really, real large earthquakes where it basically knocks almost
everything out, you know, all -- enough is knocked out so that -- and,
by the way, some are recoverable, but it's just that things break.
And those are, you know, these real rare events. But
there's another piece; there is a piece where I will call -- they're not
10-6 earthquakes, they're 10-3 or 10-4 earthquakes. They're still
infrequent, but they're not 10-6 earthquakes, in which you get a 10-3 or
10-4 earthquake, and what causes it is the failure of something else.
If there are two failures of something else, some of them
are non-seismic failures. For example, and a crucial one, is
non-seismic failures of containment isolation, and the second is,
seismic containment isolation, all right?
That seismic loss of containment isolation leads to the
LERF, because you're open, and you know you have your core melt, but
you're -- so in order to make sure that that was not a big, big concern,
in the IPEEE -- I'm proud of having been part of making sure that got
done -- we -- and I was here helping the staff at the time --
We wrote guidance to make sure that every plant did a
specific evaluation of the seismic capacity of containment isolation.
Does everybody remember?
That was the one thing we asked them to do in containment,
separate from the rest. And to our delight, actually, the seismic
capacity experts who were telling us this, told us that, but, you know,
it was very strong.
What we found wasn't a single plant in which that was a
problem. That is containment isolation, the valves, you know, they
turned out to be extremely robust.
People were telling us that, but we found it. Nobody -- no
plant that I can remember found a seismic leak of containment isolation
And that then provides you with the additional confidence
that for those infrequent initiators, the contained fission product, you
know, isn't really what I will call the common cause part.
There is still the other part, you know, which is that
earthquake you've got going by the earthquake, but then the rest of it
is an accident, you know, just the usual stuff that happens in an
accident -- the fact that the earthquake occurred 12 hours ago isn't
really what's driving the rest of that.
So that .1, you know, for the contained, is because of the
rest of it, not because of the earthquake, and that's a very important
thing that we've learned from these analyses.
MR. HOLAHAN: There is an analogous thing in fires that
we've found; that the risks are either driven by the very big fire, or a
smaller fire when other things are out of service for other reasons.
MR. BUDNITZ: You mean, a non-fire failure?
MR. HOLAHAN: Yes, right. Now, for CDF, as opposed to LERF,
about half of the seismic CDFs are seismic and non-seismic combinations,
and the other half are all sesimics.
But for LERF, they're dominated by something else; for LERF,
they're dominated by these large, all-seismic failures, and some of it
is seismic -- is random failures of containment isolation.
MR. APOSTOLAKIS: Just to move on, how can we convey the
thought that when we say .1, we really don't mean .1? It's not a speed
MR. KING: These are guidelines. I mean, this is not
intended to be a risk-based application.
MR. APOSTOLAKIS: I understand that. But if it really has
an excellent containment, modern and so on, and they say, look, mine is
really .4, would you let them raise the 10-4 to 10-3 in response to core
Is an order of magnitude too much, in other words?
MR. HOLAHAN: The answer is no. Give me a harder question.
MR. APOSTOLAKIS: I don't know why you would say no. I
mean, one in a thousand is not --
MR. HOLAHAN: For core melt?
MR. APOSTOLAKIS: I think that comes back to the discussion
this morning that it's not just that you're trying to optimize, you
really don't want to see core damage.
MR. HOLAHAN: Right, exactly. Yes.
MR. KRESS: There is some floor on core damage.
MR. APOSTOLAKIS: How do we send that message that maybe a
MR. HOLAHAN: We have a subsidiary numerical objective of --
MR. APOSTOLAKIS: These are supposed to be a means, mean
MR. HOLAHAN: Yes, of 10-4 for core damage frequency, and we
have a safety goal that says prevention of core damage is one of our
MR. APOSTOLAKIS: If we put this in a diagram form and put
shades of gray --
MR. APOSTOLAKIS: This is really misleading, .1. Actually,
we're going to the three-region regulatory scheme where there is an
unacceptable region, we talk about between that and the goal, and then
MR. HOLAHAN: That sounds like a speed limit.
MR. APOSTOLAKIS: Variability -- no, for the
unacceptability, yes. Oh, I bet they're going to give you a speed
Anybody who comes in here with a core damage frequency of
5-10-3, will be arrested. There is a speed limit.
MR. GARRICK: Is there a limitation on --
MR. HOLAHAN: If the term, arrested, means stop their
actions, that's probably correct, yes.
Is there a limitation on the distribution, as well as on the
MR. APOSTOLAKIS: Not yet, not yet. They only have the mean
value. I know you guys have thought about -
MR. HOLAHAN: I think if you let Tom finish the discussion,
you'll find out that you're most likely not going to find these numbers
in the regulation.
MR. APOSTOLAKIS: No, no.
MR. KING: These will result in some deterministic
MR. HOLAHAN: Right.
MR. KING: The way I envision this will be applied is that
you will take each initiator and you go through and you look at, you
know, given the system that's there are systems that are there, giving
the initiating event, concurrently -- these are sort of aggregate
When you add them all up, you want to make sure you've got
the 10-4 CDF -- minus fifth -- LERF, and I wouldn't propose we require
each one to meet a tenth of that, so there could be some flexibility.
Maybe some would meet it very well, and some would be a
little higher. But when you add them all up, you want to have the
aggregate come out to the 10-4, 10-5.
If you go through and you find out the regulations today
don't assure that you can meet these kinds of numbers, that's when I
think you come in and start looking at, do I need additional redundancy,
diversity, you know, additional QA, additional inservice inspections,
inservice testing, EQ, whatever it is to increase the reliability.
And that sort of gets to the --
MR. HOLAHAN: Before you leave this, I think this is a good
exercise. Conceptually, I've gone into this with the expectation that
if you look at the way the requirements were written in the first place,
if there were credible events, whether it was one a year or one in a
million years, we required multiple gold-plated systems to deal with it.
The natural consequence of that is, we provided too much
protection for the relatively rare events, and not enough protection for
the frequent events, okay?
And so, you know, my expectation is that when it comes to
large loca plus loss of offsite power, and these relatively rare things,
you know, we have too many requirements.
When you look at things like reactor scram and aux
feedwater, you have to make sure that you have enough, okay?
And that's generally what I think this is going to -- this
sort of analysis is going to lead to.
MR. APOSTOLAKIS: I would suggest, Tom, that given the
discussion of a few minutes ago, in addition to a goal, given upper
limits, I think it's important information.
And, again, the upper limit can be interpreted the same way
the goal is interpreted, not as a crisp line, but --
MR. KING: You mean an upper limit like this?
MR. APOSTOLAKIS: No, no, that's on a different quantity.
Let's go back to the previous one.
MR. KING: That's on the total.
MR. APOSTOLAKIS: Like let's you talk about anticipated
initiators. My goal is for the event response to prevent core damage of
a 10-4 number.
But anything above 10-3 is unacceptable, too. Two numbers
instead of one, in other words. Because that's the reality today, and I
don't see why we can't reflect reality there.
And if you have a problem with interpretation of 10-3, I
suggest you have the same problem with the 10-4. So these numbers
should not be interpreted as being absolute speed limits.
But at least you send the message, and I think this idea of
an acceptable, tolerable, and don't care regions, is a good one.
MR. KING: I understand what you're saying. I'm not sure --
MR. APOSTOLAKIS: Whether it's 10-3 or something else, I
don't know. That's what we just threw out.
MR. KING: Clearly, if we were going to apply this in a
mandatory fashion to existing plants, what you said would probably have
to be done. But remember, this is a voluntary program.
MR. APOSTOLAKIS: Sure, but even in a voluntary situation,
or even guidelines, it helps to give to guidelines as much as you can,
so people know where they stand.
I mean, the truth of the matter is that the core damage
frequency right now, greater than 10-3 starts also some valid --
management of the attention and so on. And yet we don't say that
anywhere, we just act that way.
What I'm saying is, why don't we say it someplace? If you
have a goal of 10-4 for core damage frequency, but we don't say
anywhere, what we really do.
What we really do is we allow 19 units to be above the goal
and we do nothing, but if anyone comes in here with a calculation that
the core damage frequency is greater than 10-3, things do happen.
MR. KING: Remember what we're trying to do in Option 3;
we're trying to come up with some revised regulations and if the plant
volunteers to meet those, they will now have to have system structures
and components and an operation that does bring them in at 10-4, not
MR. APOSTOLAKIS: I understand that, but what I'm saying is,
you will be giving them a more concrete guidance if you follow that
approach, because you're telling them, really what you expect them to
And that's something to think about, or maybe Joe Murphy can
think about it.
MR. KRESS: Let me ask one more question about this table.
If you look at the conditional containment failure probability line, I
contend the lower that number gets, smaller the uncertainty is in the
Do you reach a limit of the uncertainty in the bypass, but
you get rid of all the other uncertainties to the failure, early failure
in the mode and the location.
And if then they got down to a level of .01 instead of .1, I
think you're near that minimum in uncertainty in the LERF.
It seems to me like that's a desirable -- since the
defense-in-depth is to deal with uncertainties, unknown and known, it
seems to me like having that uncertainty at a minimum level would be a
desirable thing to shoot for.
MR. KING: I'm not sure why you say the uncertainty would go
down. I mean, you still may have a wide band of uncertainty about it,
even though it's small.
MR. KRESS: It would be minimum. I don't know how big it
would be, because you get rid of the uncertainties due to the failure --
design versus failure location, the location of the containment.
As that conditional containment failure goes down, it means
you've got a bigger, stronger containment with more reliable systems.
MR. KING: You get rid of scenarios that lead to failure.
MR. KRESS: Get rid of all the scenarios that lead to
failure, except the bypass.
MR. KING: But the ones that are left, well, if it's just
bypass, yes, that --
MR. KRESS: Yes, so I'm saying there is some reason to make
that number smaller, and that is because it minimizes the uncertainty in
And I don't know if that's -- I just thought I'd throw that
out as a concept.
MR. KING: I hadn't thought about it.
MR. APOSTOLAKIS: Did you say you will think about it?
MR. KING: I said I had not thought about that aspect of it.
MR. KRESS: That was in my talk this morning. That was the
MR. APOSTOLAKIS: Did you reject my suggestion, or you will
think about it?
MR. KING: I'll think about it.
MR. APOSTOLAKIS: Good.
MR. HOLAHAN: I believe he's thinking about it right now.
MR. KING: All right, I think we talked about most of this.
We would use mean values.
In the table we show that numbers is associated with full
power, but we'd also apply this similar concept to the shutdown
condition as well.
And then my last slide, okay, what do we do with this
working definition? As I said, the idea was to take each initiating
event and follow it through to see if you can meet those risk goals or
what you need to do to meet the risk goals.
We're also going to take a top-down look where you take
these four cornerstones and line up today, what's in the regulations,
Reg Guides, SRPs, under each of those and take a look at the balance in
terms of there are probably a lot of things that affect reliability and
availability and redundancy and diversity of systems to respond to
Do we need similar types of requirements when you talk about
containment? Is there more we should do under prevention? What's the
balance when you come down vertically at each of the cornerstones?
So, that's sort of the concept that we're going to apply in
the application of that table.
Again, I just want to say that in terms of wrapup, what
we're talking about this is the basis for looking at the regulations.
We're not talking about putting these numbers into regulations; we're
talking about using these to come up with some change in the
And we're not talking about putting in the regulations, a
rule or a definition of defense-in-depth. I think it's a philosophy
behind everything that's going to end up going into the rules.
MR. KRESS: I think the table itself is almost a definition.
MR. APOSTOLAKIS: Yes. Okay.
MR. KRESS: I like the approach myself. It's pretty much
what I was advocating this morning, I think.
MR. APOSTOLAKIS: This is the pragmatic approach. Very
MR. KRESS: Very good. We appreciate that very much.
MR. APOSTOLAKIS: Based on what we saw today, it's very
MR. KRESS: I don't know how we'll apply that to Yucca
Mountain, but --
MR. APOSTOLAKIS: The staff refuses to take it seriously,
but maybe one of these years.
MR. KRESS: Well, it's a way to handle uncertainty. I'm not
sure how we apply this to Yucca Mountain, but --
MR. APOSTOLAKIS: I think it's a different beast.
MR. KRESS: I think it is, too.
MR. APOSTOLAKIS: I think the fundamental difference is
time, the time scale.
MR. KRESS: We're due for another break. Does anybody need
MR. APOSTOLAKIS: Yes, we do.
MR. KRESS: Another 15-minute break.
MR. KRESS: The next item on the agenda is to hear some
words from the NEI and the industry, and from Westinghouse, so I'll turn
the floor over to you, Alex, and let you introduce the subject and
introduce the people.
MR. MARION: Good afternoon. My name is Alex Marion, and
I'm the Director of Programs at the Nuclear Energy Institute. I
recognize the time is late, but I do have a few brief comments to talk
about some of the things I heard today relative to the application of
defense-in-depth philosophy to operating plants.
But I would like to introduce Rodney McCollum, who is the
Project Manager at NEI involved with high level waste management, and he
has a few comments he would like to make on the application of that
philosophy to the Yucca Mountain Project.
MR. McCOLLUM: Do you want me to go ahead and do that first?
MR. MARION: Yes, please.
MR. McCOLLUM: I've been working for NEI now for a little
more than a year, specifically to follow Yucca Mountain and related
issues, so I have been attending meetings such as this one, and hearing
discussions such as I heard today for most of that time.
I always find these discussions very interesting and very
intellectually challenging. I think this one was definitely no
exception and perhaps even a little bit too much so on the
intellectually challenging part, but that's how I learn things.
I also feel it's a very important discussion, and it's
certainly a very timely discussion because the nation is entering into a
critical window of decisionmaking opportunity here where over the next
18 months, our leaders are going to be called upon to make a decision
about the future of Yucca Mountain.
And one of the things that will weigh most heavily in that
decisionmaking process is the topic of uncertainty that's been discussed
a lot today.
How will the decisionmakers, relying on the Nuclear
Regulatory Commission, the ACNW, the TRB, and all of the political
forces that come to bear, how will they view uncertainty?
And uncertainties will exist; that's really the only thing
that is certain. In fact, if it's good enough science, every answer
will simply generate more questions, it will bring up more
And, therefore, because these uncertainties will inevitably
exist, the decisionmakers need to have some tools in place that will
allow them to address this.
And we firmly believe that the DOE, in the viability
assessment, and the NRC in the draft Part 63, is giving them these
tools. We feel that in referring to what Christiana was talking about
earlier, the way multiple barriers are being interpreted, that it is a
qualitative and not a quantitative argument, and that it should be up to
DOE to make the safety case. We feel that's appropriate.
We are concerned to the extent to which at this point,
having seen what's been done by both the DOE and the NRC staff to
develop those tools, what could be gained by inserting knowledge on the
reactor side from the reactor notion of defense-in-depth into the
We've had a lot of discussions along this line with our
friends in EPRI included, and perhaps the best way for me to relate what
might happen if we were to bring these things in is:
I, once a upon a time, was a Branch Chief of Nuclear Safety
for a DOE operations office that had responsibility for a lot of very
unique, one-of-a-kind, non-reactor nuclear facilities. We had a couple
small reactors. This was the Chicago operations office so we're talking
about the Brookhaven's, the Argon's , the Princeton's, et cetera.
And I was in that position at a time when DOE was coming out
of its post-Cold War cocoon of beginning to realizer that it needed to
have some credible nuclear safety requirements, a regulatory structure
in place that it didn't have before when it simply did what it knew was
right or thought was right.
And doing son, they naturally looked to the best source of
expertise for that kind of a regulatory structure, and that was the NRC.
So the DOE made a lot of requirements that were first under the guise of
DOE orders, and later became -- a couple of them became rules, DOE
rules, that basically took NRC regs that were intended for the reactor
world, and put DOE order numbers on them and they were to be applied to
these non-reactor nuclear facilities.
Once that happened, I found myself spending a lot of time
trying to fit square pegs into round holes, and trying to explain why
the square pegs wouldn't go in the round holes. That they just don't
fit, never quite seem to be enough of an answer.
I saw a lot of effort being made at the five National
Laboratories to address all those misfitting pegs that didn't contribute
to their safety cases, and, in fact, just detracted from it.
I was very appreciative to hear what Dr. Garrick said
earlier about arbitrary thresholds and subsystem requirements that
detract focus from risk. I know from experience that that that does,
And I think we have a pretty similar situation here with
Yucca Mountain, because Yucca Mountain would be a very unique,
one-of-a-kind, non-reactor nuclear facility.
I think that the differences between Yucca Mountain and
reactors are so fundamental, it really becomes almost impossible to try
and draw from reactor defense-in-depth to multiple barriers in the
A couple of those things have been mentioned, a couple of
others, I would mention: Of course, obviously you have more active and
passive barriers at Yucca Mountain, whereas you have more active,
engineered and more engineered features at a reactor.
Yucca Mountain has one common failure mode, really, a
two-part failure mode. It's water and time. And it's really a question
of where you are on the radioactive decay curve when those things attack
each of your barriers.
There are different timeframes to be considered. In
reactors, fractions of a second can be important; in repositories,
millennia are what's important.
You have a safety case in reactors where you're trying to
figure out where to best apply PRA; in a repository, your safety case is
You rely on humans to operate reactors; your expectation for
the repository is that once you seal it up, except for potential human
intrusion, humans won't be involved at all.
And probably the most important distinction that allows you
to treat uncertainty in a fundamentally different way at a repository
would be that you have this performance confirmation period. You have
not a two, but a three-stage licensing process.
And this is a 50-year period where you have a chance to
constructively address those what-if-we-were-wrong questions.
You don't have that at a reactor, and I don't think any
utility would want that, although some time felt they were approaching
But it does give you an opportunity, and it does give the
decisionmakers to say, when they are faced with uncertainties, that
here's what we know, and here's what what we know tells us, and then
here's what we need to know before we close the thing, and put in place
the right research program that can answer those questions.
But you can't do that in the reactor world. So, given that,
and having heard the discussions -- and this is another one of the
things we appreciate where the staff is going with multiple barriers. I
was very thankful to see Christiana's presentation entitled Multiple
Barriers and not defense-in-depth.
We wonder -- and this is kind of the conclusion of the
discussions we had internally -- whether defense-in-depth is even an
appropriate term; whether it would be more appropriate to call what
you're doing at Yucca Mountain multiple barriers and call what you're
doing in the reactor world, defense-in-depth, and not even try to mix
It could only lead to a confusion in expectations, and as I
mentioned before, you know, we think the expectations are evolving well
for Yucca Mountain. We think that Part 63 will answer that.
We think that from what we've seen at EA, and from DOE's
draft Environmental Impact Statement, they'll be able to say that when,
you know, the final dose, if it's 1.3 millirems, 10,000 years from now
or whatever it is, that that dose is a function of a performance
assessment that includes a dry climate and includes a thousand feet of
rock to keep the water out of the repository.
It includes a lot of things in the repository, some of which
are engineered, and includes another thousand feet between the
repository and the water, and it includes things in the water that
retard the movement of radionuclides.
And it includes a sparsely populated area that keeps people
away from even those moving radionuclides. And, of course, the DOe will
have looked at a certain amount of variations and been cautious and
reasonable in looking at each one of those barriers. It will assume a
somewhat wetter climate. It won't take credit for the features of the
rocks that it doesn't understand as well as it understands some others.
When I visit the folks -- and in the year, I've had three
tours of Yucca Mountain now, and I talked to the scientists in the
tunnels and hear them talk. I appreciate what Dr. Levinson mentioned
about some uncertainties are not bad.
They are tending to find out things about the rocks that are
good news. And they will do that during the performance confirmation
But based on what they know, they can make a case that that
1.3 millirems or 13.2 millirems, or whatever number it is less than 15
or 25, is a function of a number of things. And those things all
contribute to it.
And in that respect, it need not be much more complicated
than that. They will have then answered what Congress has asked for in
terms of multiple barriers, and the NRC can and should, in accordance
with its regulations, look very hard at that and make sure it's
credible, that it's believable before the Commission says to the
decisionmakers, we think this is sufficient, which is the sufficiency
comment component of the site recommendation.
Then we go on to the next stages in the process, and we
continue to look at it, realizing that the scientists will never stop
asking questions, and that every one of those questions will bring into
the proces, more uncertainties, and that's not a bad thing.
So, you know, I'm very encouraged that these discussions are
occurring, and I learned a lot from them, and look forward to this going
MR. MARION: Are there any questions of Rodney before I make
a couple of comments?
MR. APOSTOLAKIS: I don't so much have a problem with the
regulations, the way Christiana presented them. It's really the quality
of the performance assessment that would be of concern to me, given the
time scales we're talking about and the uncertainties that are involved.
And I still don't believe that the model uncertainties are
completely addressed. Even in WHIP, you know, there was primarily
parameter uncertainties. At one point they had two different models for
something relatively minor. I don't remember what it was.
They said, okay, we'll put a weighting factor of 1/3 to
this, and 2/3 to the other, and just add them up. But I think the
uncertainty is a key issue here.
MR. McCOLLUM: Oh, they clearly are. As I mentioned,
they'll be the major thing weighing on the decisionmakers.
And that is why, in demonstrating multiple barriers, DOE
needs to talk about what each of those barriers mean to the safety case,
and what is the meaning of those uncertainties?
And they're starting. And every time I have heard DOE
present on this subject now, dozens of time, and the story gets better
every time, that the science was always there, I believe. It's been
there since the VA.
But it's being able to talk, and it can't be completely
quantified. It shouldn't be. But to be able to talk about the relative
importance, what does that uncertainty mean, what if the climate does
get wetter? Have we looked at that?
Have we been appropriately cautious in what we've assumed
the rocks do for us, and what we've assumed the rocks don't do for us?
And so that if some of those uncertainties turn out to be
bad, are there offsetting things? And it's really going to be a
challenge in the next 18 months when we have this decision before us,
for that to be discussed.
And I have also heard Dr. Garrick talk a lot about plain
english, and that's why that's so important. Because those things may
be buried in the performance assessment in any number of ways, but if we
can't bring them out and discuss them in plain english so people
understand that that's what this means, that's what that means.
And because we know what all these things mean to the safety
case, we can say this is a good place for a repository or not. And we
can make a decision.
MR. GARRICK: George, I think that the Committee kind of
shares the concern for the TSPA. We know that in the early days of the
PA for WHIP, there were many, many problems, and through another
Committee, I was directly involved in that.
And I saw a major change. The big difference there over
Yucca Mountain is that except for human intrusion, there was geologic
containment at WHIP.
And the only way WHIP could get in trouble was through some
rather arbitrary human intrusion scenarios. Of course, we don't have
that luxury on Yucca Mountain.
MR. APOSTOLAKIS: Right. The other thing that we did that
you guys may find disturbing is that later on, I believe, 60 hypercube
simulations. All 60 of them were below the goal, which brings us back
to your comment, what if it is 5X?
What if Yucca Mountain, 58 of them are below and two are
above? That will create an interesting interpretation of the
And why should all 60 be below? Just because it happened
Now if you think of the state of knowledge on uncertainty,
the whole distribution is below -- I mean, the two high percentile, so
that -- anyway, these are not directly related to the subject matter.
MR. GARRICK: It's a good comment.
MR. MARION: Thank you. I'd like to make a couple of
comments about the operating reactor side.
I found Dr. Murley's comments this morning kind of
interesting. Having worked at a nuclear utility for 15 years, it sure
felt like defense-in-depth was a regulatory requirement at times.
MR. MARION: But I decided not to challenge it.
MR. APOSTOLAKIS: It was a voluntary requirement. We have a
lot of those.
MR. MARION: But I thought he made an interesting comment
about -- or a caution, I should say, as I interpreted it, about applying
risk insights to remove or otherwise eliminate barriers.
I think we need to be very careful, and I think that's an
appropriate cautionary statement. However, I think with risk insights
and operating experience, we can better define what's important in the
implementation of the very elements, specific elements of those various
barriers of protection, specifically in the area of emergency planning.
I believe we're very close to the point of providing a case
to reduce the exclusion zone, based upon the robustness of the designs,
as well as the analysis supporting the advanced reactors.
And there are opportunities. We're not offering to get rid
of emergency planning as a concept, but better define it with the latest
intelligence and knowledge base we have.
And I think that's consistent with the comment that Dr.
Budnitz made about the evolution of knowledge to better focus on
barriers of protection, integrating operating experience and new
And I think we need to keep that in mind and take advantage
of those kinds of opportunities when they present themselves.
I think the example that Dr. Apostolakis used on the fire
analysis and the element of smoke and uncertainty associated with it is
an excellent one in terms of applying an engineered approach to address
And then when knowledge comes to bear and the analytical
techniques improve to better reduce the uncertainty in the area of smoke
propagation, et cetera, then you can make adjustments along the way.
And I think those were excellent examples, and we're in full
agreement with those concepts and processes. And in NRC staff's
presentation this afternoon, I was sitting back there with Biff
Bradley's, the project manager at NEI directly involved in
risk-informing Part 50 and these PRA risk insights, applications, et
And he indicated to me that we're in full agreement with the
approaches. And I think, between the industry and the NRC, we're in a
good position where we understand the importance of striking a balance
between the deterministic thinking that's made this industry very
successful within the defense-in-depth philosophy, and applying that in
some balanced way with probabilistic techniques and approaches that we
And from what everybody tells me, things are going well in
terms of the applications of risk-informed regulations, but we do have a
lot of work ahead of us.
And I just want to caution everybody that we want to be
careful not to limit our thinking or limit our approaches such that when
new knowledge or when new analytical techniques come to bear at some
time in the future, we can still take advantage of those and improve our
knowledge and understanding.
This is the defense-in-depth philosophy balanced with
risk-informed approaches, and is very fundamental to our thinking for
regulatory reform, more specifically in the area of risk-informing the
Part 50 regulations.
So we think it's very important to work hand-in-hand,
shoulder-to-shoulder, so to speak, in a complementary way with the NRC
staff, and to strike this balance and determine what we need to do with
future applications of the current state of knowledge.
And that completes the comments that I have. Are there any
questions about anything I said about operating plants, or that Rodney
MR. MARION: Okay, with that, I'd like to introduce Gary
Vine from EPRI, who is going to take a few minutes and provide you with
a general overview of the defense-in-depth philosophy as it was applied
in the design requirements for advanced reactors.
I think you will find that informative and beneficial. And
he will be followed by Brian McIntyre from Westinghouse, who is going to
specifically discuss the application of that philosophy in the AP-600
MR. APOSTOLAKIS: One of the victims of defense-in-depth.
MR. MARION: We were going to bring that up a little later,
MR. APOSTOLAKIS: Perhaps the only one still alive.
MR. APOSTOLAKIS: While these are getting settled, somebody
said this morning that there may be a perception out there that we're
using risk-informed regulatory approaches to remove barriers, to remove
regulations and requirements.
I think it's important to say that where PRA indicated that
additional requirements were needed, the Agency acted immediately. And
in the last 20 years, in fact, the eagerness of the Agency to add
requirements based on PRA insights created a somewhat hostile view
within the industry towards PRA, because PRA was used only to add
So the fact that now we are finally looking at removing
some, should not be misconstrued as the Agency using PRA to remove
requirements. We have already added a lot, okay. That's in case
anybody reads the transcript.
MR. KRESS: Thank you, George, I think that was well said.
MR. VINE: Good afternoon. I'm going to start off. My name
is Gary Vine. I'm from EPRI. Unfortunately, I didn't have the benefit
that Alex and Rodney and Brian did of all the prior discussions. I got
here about 4:00 from another meeting in Tower I.
But Alex does tell me that a number of the points that I
intended to cover have been covered in some way, and so I'm going to try
to focus only on either new material or kind of an industry perspective
on some of the things you have heard from the NRC side.
I'm going to probably skip over the first slide or two. The
only key point on the first slide is simply that we did in the ALWR
program, which goes back 10-15 years now, fully embrace the concept of
And we did that in a traditional way. I think we didn't use
the terms that you've been discussing today, structuralist and
rationalist models, but we pretty much followed the traditional
I also have a slide on ALWR policy statements, and I
intended to go through two or three of them in some detail, and I'm
going to skip that as well.
I have a high-level brochure document that provides a two-
or three-sentence description of each of these policies, some of which
have a bearing on defense-in-depth, and I'll just leave that for you to
Moving on to Slide 4, just a couple of key points: It's
very important to recognize that public health and safety is important
to both the NRC and to the owner/operator of a plant. In fact, the
owner/operator has the primary responsibility of protecting public
health and safety.
So his interest in safety is just as high as that of the
regulatory. Where the difference lies in the way we fundamentally
approached establishing design requirements for advanced reactors is in
the investment protection side.
That is where the industry has an equally high interest in
preserving their investment. But the NRC doesn't have a comparable
And so what that forced us to do was to make a lot of
tradeoffs as we were trying to optimize prevention mitigation
decisionmaking where the industry's interest was naturally always to
achieve safety as early in a sequence as possible.
We always wanted to prevent an accident or actually have a
robust enough design so that we wouldn't even get into an accident
sequence before we had to get into questions of mitigation.
We also found when we had a fresh sheet of paper and we
could look at these decisions, that almost always -- not always, but
almost always, when you had a particular sequence you were trying to
drive down or improve the safety for and you had a mitigation option and
a prevention option to do that with, the prevention option was usually
So there were a lot of incentives on the industry side to
truly tackle areas where we wanted to achieve improved safety by doing
it on the prevention side. Of course, this, as you can tell, created
some friction between the industry and the NRC, on occasion on certain
issues where the thought was that we were maybe not maintaining the
proper balance in defense-in-depth.
We maintained a strong commitment to mitigation as well.
Requirements for containment, for example, are just as strong or
stronger for advanced reactors than they are for current plants.
But as we pressed to achieve improvements on the prevention
side, there came some questions about balance.
Explicit consideration of severe accidents via a safety
margin basis, that's a very important concept which I think is probably
worth some discussion. I think there were some understandings in kind
of a process way in the program with the NRC that have stood the test of
We fundamentally committed to the licensing design basis as
it was captured in Part 50. And we did not, with just a very few
exceptions, try to make any changes to the regulations.
The only example on this schematic where we tried to make
some improvements in the regulatory basis in the licensing design basis
side was in improving the source term that was analyzed in the licensing
But we pretty much bought into the entire licensing design
basis approach as, quote, the "formal speed limit" for design.
But we were very careful in defining very separate and
distinct from that licensing design basis, the way we would approach all
other safety questions and primarily all questions associated with
In this area, there were differences in almost every aspect.
We approached it, first of all, from a standpoint of a much more
risk-informed evaluation of the plant's overall performance.
Second, we insisted that we use best estimate analysis
methods, models, and so forth in addressing those issues.
Third, we proposed and the NRC accepted, the concept of the
industry pretty much driving the specific design approaches to address
severe accidents, and get the NRC to provide an overall approval to the
approach that we took, as opposed to agreeing on detailed prescriptive
requirements that would then become part of the licensing design basis
or some formal regulatory requirement for this right side of the
So the industry really drove this. We decided how we wanted
to satisfy the Commission's concerns about severe accidents, all the
research findings, the Commission policy statements and everything else.
The NRC then provided an SER on these utility requirements,
and then the vendors had a clear picture of how they had to achieve
basically what they had to do to know that they would have regulatory
approval in this area.
There were a number of areas, even though we pretty much
approached things in a conventional way with regard to defense-in-depth,
where we kind of pushed the envelope, and what I'm going to cover now
are some areas where I suppose if you get to the definitions you're
using now, where we used a more rationalist model approach or a more
risk-informed approach to the way we did business.
First of all, let me jump back to Slide -- yes, this is the
right slide. I'm sorry.
Major alliance on PRA and the process: It drove our side,
the industry side, very significantly. We made major plant policy
decisions and major plant design decisions based on findings of the PRA.
The regulatory side used PRA much more just as a
confirmatory tool, as opposed to a decisionmaking tool. One exception
which Brian will get into is the way we dealt with the regulatory
treatment of non-safety systems for the passive plants.
But beyond that, the regulatory side was pretty much a
confirmatory process. We established quantitative safety requirements
on the industry side that well exceeded the regulatory requirements.
And the idea here was that we wanted assured license ability
by knowing we had significantly exceeded what the regulatory
requirements were going to be in the area of safety.
I list our two quantitative safety requirements, and these
were requirements; they weren't just targets: The designers had to have
a CDF much less than 10-5, and they had to address mitigation by meeting
a goal of ensuring that whole-body dose would be less than 25 rem at the
site boundary which is about at a half mile as we defined it for all
sequences with a cumulative frequency of greater than 10-6.
You will notice that these two prevention and mitigation
goals are not coupled; they are decoupled, which gets to my final point
on that slide:
We did oppose the concept of coupling these independent
layers of defense-in-depth. We opposed the concept of a CCFP. We
didn't win that argument, but we do believe that CCFP is not an
appropriate means of enforcing a defense-in-depth approach because it
couples things that should remain independent.
Because one is set by design, you end up forcing the
operator or the designer to make less than optimum, sometimes dumb
decisions in having to reduce the safety of the plant in order to
maintain this spread of a factor of ten between prevention and
And there are -- you can go through some scenarios down on
the low probability events where the imposition of a CCFP becomes even
So we felt that that was an inappropriate approach and still
Regulatory stabilization: I already mentioned assured
licensability by exceeding the regulations wherever feasible. This was
an important concept to us, and we've faced some problems in dealing
with the NRC on this because we wanted to assure significant and visible
and demonstrable margin between the regulatory requirements and actual
design performance and operational performance.
And there is just a natural tendency on the part of the
regulator to say, well, gee, since you're that much better, let's just
change the speed limit so we're a lot closer to where you are.
Well, that creates huge problems for us, because it
eliminates that assured licensability. And so we think that the
regulatory requirements ought to be based on the first principle and the
bases upon which NRC makes its regulations on adequate protection and so
forth, and allow the user of those regulations to exceed them and not
have that difference gobbled up into regulation.
There were a few case where we attempted to change the
regulations. We would propose in some areas -- these are usually some
modest areas -- we didn't go after things like large break loca and so
We did propose some changes to the regulations, and some of
them were accepted and some of them were not. Some examples that were
talked about were: More realistic source term; elimination of the
operating basis earthquake and going only with the safe shutdown
earthquake; changes to hydrogen regulatory requirements.
This optimized or simplified emergency planning that Alex
mentioned earlier, and so forth.
And the last slide I think is more just personal views as we
look back over the ALWR program and how we approached defense-in-depth.
We think that looking forward, that risk-informed regulation and
specifically a more -- an approach to defense-in-depth that is closer to
the rationalist model is really important to the future.
We are going to have to find ways to reduce the capital
costs of ALWRs, and we believe that can be done easily and safely, and,
in fact, probably in many ways improve safety.
But it does require more flexibility on the regulatory side,
and a rationalist approach would allow for that.
Further, I don't see how the NRC will ever be able to
license a reactor design such as a high-temperature gas reactor, unless
there is a more flexible approach to defense-in-depth, including
something similar to the way you've characterized this rationalist
I think the die is cast; the rationalist model is ultimately
going to become the future approach for regulation, and I don't think we
need to be afraid of that. I think there are really no downsides to
that model, if, in fact, it's done prudently and carefully and safely,
and done with the things that are already pretty much established in
regulatory policy, namely, that it's not going to be a risk-based
approach; it's going to be a risk-informed approach.
There will be a balance, there will be still consideration
of defense-in-depth, there will be clear use of engineering judgment and
care and so forth in how you approach risk insights.
And just finally one comment on U.S. leadership: The ACRS
paper on defense-in-depth mentions a couple of INSAG reports, and it's
true that in the international arena, there is a much more rigorous
definition, a much more traditional and formal approach to
And I think there probably will be some resistance on moving
quickly toward, say, a rationalist model, internationally, and the
reason is that I think there is a concern by IAEA and probably some of
the industrialized world regulators that if you move too quickly, you're
going to find some countries, third-world countries, people who don't
have the maturity and infrastructure, safety culture, and so forth, that
if you move to quickly in optimizing defense-in-depth philosophies, that
you're going to remove some significant safety protection.
And so there will be some desire, I think, in the
international community to move slowly and to make sure that, especially
for those who define defense-in-depth very broadly -- and I've seen it
defined this way to include things like safety culture and your
infrastructure and your regulatory infrastructure and so forth -- that
those things still are not subsumed under a risk approach, and you don't
make them subservient, but you still keep them at a high level.
MR. APOSTOLAKIS: It's important, of course, to note that
terms like quickly and slowly are relative.
MR. VINE: Yes.
MR. APOSTOLAKIS: And that the first major risk assessment
in the United States was published a quarter of a century ago. So for
us, it's not too quickly.
MR. McINTYRE: My name is Brian McIntyre, and I'm the AP-600
License Manager. I'm two things: I'm the practical application of what
Gary just talked about; and I'm also, I think, the most recent example
of where the rubber has met the road with the staff on defense-in-depth.
And this is -- we have really talked at lot about this, I
think, earlier, that it's more than the three barriers that was
originally put in to deal with uncertainties.
What I had written down is that we are never sure exactly
what it was. And after sitting through this morning, I think it's that
everybody was more or less sure what it was, and it was whatever it
needed to be, and it was sort of a flag that we all wrapped ourselves
in, both sides, I mean, the industry and the regulators.
But we never quite knew when enough was enough, and I'll
talk about that at the very end of this. And now it's clear that we are
moving towards some sort of a balance between the things that are on the
top there and the risk-informed information.
In the AP-600 case, for us, I broke this down into two
things, something that I called the unquantifiable aspects -- and this
goes beyond just power reactors. For us, it was a design philosophy.
Now, at the bottom I have some things that are quantifiable.
We actually, since were starting from scratch, weren't
trying to figure out how good the plant was; we were more interested in
how good we could make the plant. And you really take a different
approach if that's what you're doing.
And our design philosophy looked at -- people have kind of
wondered about passive plants -- that we have multiple levels of
And the first thing that you see there is that it was
usually a non-safety, active feature. We have a passive plant, and that
made the staff -- these are my words -- made them a little bit crazy.
Because, as you're going to try to address your transients
by using non-safety systems, this is as a first shot, yes. And then
almost the backups would be the passive systems which were the safety
And if you want to look at what this looks like, the next
figure or the thing that actually is the figure, this is -- and we did
this for a number of transients where we went through and we looked.
On the left side is a current plant -- and I need to put my
glasses on to see this -- that what they would do, their SSAR safety
case is that they would automatically actuate their high-end safety
injection, their aux feed; they'd isolate the steam generator, and
they'd start to cool down and depressurize, and that was their safety
And if that isolated the leak, that was great, and if not,
then they had a non-safety case which would be in their emergency
operating procedures someplace, and they had a couple of things that
they could do. If not, then they were at a core damage situation.
For the AP-600, if you take a look at our top block, which
is the non-safety case, really, it's the same things that in a
traditional plant would be their SSAR safety case, except we had now
made these systems non-safety-related, which was really a change.
And there were some long discussions we had with the staff.
Gary talked about regulatory treatment of non-safety systems, and I'll
talk a little bit at the end about how we did approach that.
And then we got to our safety case, all these passive
features of automatically actuating the core makeup tank; the PR/HR heat
exchanger, which was basically replacing the axillary feed or startup
feed system in the safety case; the CVCS.
We'd isolate the steam generator and start the passive
containment cooling system, and if that isolated the leak, then that was
our safety case. And that's what we basically met the safety
The important thing to look at in the AP-600 is that down
below it there were then two or three other options that the guy could
go through. And this was important because, you know, we could have
just really stopped at the top, at the safety case, and with the top
For various reasons, because these features were in the
plant, that they all managed to work together, and as a result, we got
really some good PRA results. But this, to us, was what we considered
to be the defense-in-depth.
We also used the PRA as the design tool. And that's like a
lot different if you're trying to figure out how good you can make the
plant, as opposed to how good the plant is.
We did a total of seven PRAs on the AP-600. And we weren't
dong them just to make the PRA different; we were doing them because
we'd made the plant different.
We'd run the PRA, we'd find out where the weak spots were.
This is where you're looking for the unduly -- not unduly dependent on
one system, so we were looking if something really stuck out, and we'd
go back and we would make the system better.
There was a lot of design with arguments even between the
risk analysis people and the designers. We actually got better PRAs as
a result of that, because sometimes the PRA people didn't understand
exactly how the system should have worked.
In a lot of cases, the designer said, you mean that if this
fails, then that's the result you're going to get in the PRA space. And
we made some significant changes to the plant as a result of the PRA.
We went through a lot of just discussions, review,
understanding the results. We looked at some of the backup slides.
When we got to reviewing things to see how we would expect
the systems to work, this is just one example. This is the PR/HR heat
exchanger. How would it fail? We would then walk through the various
things and decide what we needed to either to try to fix or to model or
not model in the PRA.
We went through each one of the various items, for example,
for the inadequate IRWST water level, and then that was broken down to
look. Are there things that we could fix, are there things that we
needed to do better?
I mean, we really did chase this design down to look for
ways that you could improve the plant.
And this is a philosophy, so it's not just applicable to an
AP-600 or a BWR or something like that. But if you think like this and
you bring this approach to the design and bring whatever it is from a
design to actually a facility, this works.
This is another way to look at defense-in-depth, but there
is no way that we could put a specific number on what we got out of
We also looked at shutdown operations. We looked at low
power operations. We pretty much covered the waterfront.
One of the bullets on the previous slide was that for
systems that were more -- or for events that were more likely,
initiating events, we had more backups.
For steam generator tube rupture, a reasonably likely event,
there are five or six different thing you can do. When you get down to
the more unlikely things like large loca, you don't have quite as many
options of things that you can do, so we tried to focus our efforts on
the things that are more likely going to happen.
Also, one of the big reasons we were doing this is the big
push from the industry was this investment protection concept. If
something is more likely to happen, then we don't want to lose the plant
as a result.
We want to have things that the guy can do. He might have
to clean the plant up, but he won't lose the plant as a result of it.
We looked at a much wider range. We didn't restrict
ourselves to the design basis transients. We really looked at multiple
steam generator tube rupture, not willingly, but we looked at multiple
tube rupture, because this was a case of the staff's concern which was,
okay, you guys met the design requirements, but do you fall off the
And the staff went to the extent of, after we had completed
our testing at the Oregon State facility, which was a quarter scale
model of an AP-600, it was a low pressure facility, but they went out
and ran beyond design basis transients there to look to see if there was
someplace that we hadn't tested that they could look to see if we were
going to fall off the table.
And the conclusion was, no. It was a surprisingly robust
plant. I mean, we'd been telling them that for a long time, but
eventually, it became obvious.
We also looked at a broad range of initiating events. And
as I said, this was to look beyond where you would normally go.
And, again, we're trying to figure out how to make it
better, not how good it is. And it's almost like IPEE and IPEEE, except
we could make the changes, because it's quite easy really to make a
If you look at the quantifiable aspects, we ended up with
really a nice low core damage frequency. I'll talk about the focused
PRA in a second.
For large releases, what we were required to do by NEPA was
to look -- and SAMDA, if you're not familiar with those, those are
severe accident mitigation design alternatives.
I look at it as we had to explain to the staff, why we
didn't do what we didn't do. It turns out we're not really good at
documenting that, so we went through and have to figure out, why didn't
you make these changes to the plant, and you have to look at that on a
cost basis, the cost/benefit basis.
And it turns out there was nothing that we had to add,
nothing that could be cost effective when we finished the design of the
Our PRA results: This is looking at two things, the core
damage frequency and the large release frequency. It's the at-power and
the shutdown events.
The baseline PRA is pretty much a traditional PRA. It has
the safety systems and the non-safety systems in it.
As part of our ongoing discussions with the staff and the
regulatory treatment of non-safety systems, we had an approach proposed
by the industry, accepted by the staff, that if this plant was so good
that we could go out and meet the safety goals to 10-4 and 10-6, with
only the safety-related systems, then these non-safety systems that were
in that top tier or the first thing that the operator might actually do
to the plant to mitigate an accident, then they wouldn't require any
And it's a sensitivity study, but we went back and looked at
it, and we showed that without the safety systems, we still, in the core
damage frequency area, we quite handily met the safety goal. In the
large release, well, it was close.
And the staff's concern was, well, uncertainties in the PRA,
we're not so sure about this, and we went back and forth and back and
forth and back and forth and back and forth.
And finally, it just went forth, and we said, okay, to move
this forward, we would put some administrative controls on certain
systems. And so we actually have in the AP-600, safety-related,
non-safety-related, and then there are these RT&SS important systems
that we have availability controls. So we're actually --
I would actually look at this as beyond risk-informed. It's
almost risk-based, this sort of an approach that you have a milestone
that you're trying to meet, that if you do this then you will be okay,
and if not, then you'll have to do some things to make it so.
And at the time, this was quite novel. It was much for
discussion, but it certainly is, I think, a case of how defense-in-depth
can come and be played through and be applied to a facility.
One of the reasons that you're here -- and this is sort of
-- if you look at Tab 1 in Jack's book of defense-in-depth discussions,
it was that we had a long discussion with the staff on containment
spray. The AP-600 does not have a containment spray.
Well, it does have a containment spray; it didn't have a
containment spray. Let's put this in perspective and in the proper
And we didn't think that we need it, and it got back into
arguing about the uncertainties and the PRA and the models. In the end,
we ended up, as I said, with a containment spray system.
If you look at it from a risk-informed perspective, the --
and this is a slide that was put together by an ACRS fellow at the time
back in June of 1997 when this discussion was going on.
It gives you an idea of where our risk contributors are.
And for this plant, if you look at what a containment spray would help
you with, it's not going to help you with the bypass events or with the
early containment failure. It might some -- it would help you with the
containment isolation failures.
A presentation that I made to the staff had -- and you
haven't seen this one, George, but it has the more quantified basis of
what we would expect to get out of the spray.
And the spray here where it says low flow, it's lower flow
than the spray that we actually ended up putting in the plant. This was
a study that we were doing at the time to figure out how much water we
needed to make -- this is like 400 gpm, and I think we have a thousand
gpm actually in the plant.
So the spray that we have in the plant would work better
than the spray that's on this. But it shows that for earlier failure,
it would reduce it by about a factor of two, and it would help the
intermediate failures, but those are really pretty low-risk events. The
isolation failure, it would help that a fair bit.
It doesn't help the bypass, so by putting the spray in, we
ended up reducing a very small number by a factor of two. And this is
the reason that it didn't make the cut, if you will, in putting it in
the plant from the SANDA category.
And we took this actually as far as the Commission. There
was a SECY paper, and I think it's really one of the reasons that we're
here, because defense-in-depth really got down -- this was one of the
harder arguments that we have had about what is defense-in-depth?
And I'm going to read from one of the vote sheets on this
SECY, just one paragraph, because I think this answers your question
about if you pass all the requirements, would they still make you put
something in? Yes.
And the argument was that in spite of the fact that the
proposed system cannot be justified under any of the rational
decisionmaking guidelines that we have established for ourselves, the
staff would require it anyway.
The ultimate reason seems to be that it is justified to
compensate for uncertainties in how the design will behave under severe
accident conditions. Even this reason is not well supported because we
have not established a relationship between the proposed spray and the
particular uncertainties it is supposed to address.
Defense-in-depth becomes the final justification. And then
it goes on to say that the Commission and the staff should not continue
ad hoc decisionmaking indefinitely, and here we are. That's why we're
But the answer to your question is, yes. And I think that
we've perhaps moved beyond this now, and I was glad to see Gary's and
Tom's presentations. I'm not too sure, but I can probably use that to
take the spray out.
MR. McINTYRE: Since it's not a Tier I requirement.
MR. HOLAHAN: We'd have to talk about that.
MR. McINTYRE: So that's the way that defense-in-depth
actually gets applied. If you make it a way of life, almost a mantra,
you pray to it, you decide and you think like that, and it can really
result in a lot of, I think, good things in the design. That should
answer your question that you asked about five times today.
MR. KRESS: Thank you very much. I'm not so sure that if we
had had Gary's risk-informed matrix table back then, whether or not we
would have come down on the side we came down on.
MR. McINTYRE: I think what's important is that they were
looking at the balance between prevention and mitigation, because my
argument or complaint -- complaint, that's fair -- at the time was, what
are the units on this balance?
And I think there's an attempt to do that, and I certainly
MR. KRESS: That is exactly right.
MR. GARRICK: What would be much more interesting than these
point estimates, which see -- would be the PDF stacked on top of each
other for these two cases.
MR. KRESS: Yes, that was one of our problems, too. We
didn't have any of the PDFs. And all we had were point estimates, and
that made the decision much more difficult.
Had we had those, it might have been a different story.
MR. HOLAHAN: My recollection is that you didn't have them
because they were never generated.
MR. KRESS: That's right. That's why we didn't have them.
MR. APOSTOLAKIS: That's a good reason.
MR. BUDNITZ: But the difference at Yucca Mountain is a
qualitative difference about the staff behavior, I believe. See, you
were having this argument about a theoretical plant that wasn't sited or
being built anyplace in particular, in a room in an office building like
But in Yucca Mountain, it's going to be in an arena in which
the Governor, the Senators and almost the entire population of a real
state are using every political opportunity they can and every legal
opportunity they can, not only to get in the way, but to embarrass the
And the staff is acutely aware that that embarrassment has
to be avoided, if they can, and that's why they can't find themselves,
if they can avoid it, in a situation where they're backfitting a
positive decision on what would have been a negative decision by
changing their minds halfway through.
MR. KRESS: Yes.
MR. BUDNITZ: And so they really have a different dilemma
than you and the reactor staff and at that time. It's much more
difficult for them.
MR. APOSTOLAKIS: Good.
MR. KRESS: Very, very difficult. I'm going to ask if
anyone in the audience feels compelled to add anything to what they've
MR. KRESS: Seeing no rush to the front --
MR. APOSTOLAKIS: Are the experts going to be back tomorrow?
MR. KRESS: That's a good question.
MR. APOSTOLAKIS: Are they coming tomorrow?
MR. KRESS: Tomorrow, we're going to try to wrap some of
this up and see if we can reach some conclusions, and maybe spell out
what the remaining issues are, and things of that nature, and as many of
the experts as we could get would be nice.
MR. APOSTOLAKIS: So we lost Dr. Murley then?
MR. KRESS: Lost Dr. Murley.
MR. APOSTOLAKIS: Are you going to be here tomorrow,
MR. BUDNITZ: Yes.
MR. KRESS: We'll quit at precisely noon or pretty close, or
maybe even before noon, but more around there. Okay, great. The staff,
will you be here?
MR. HOLAHAN: Yes.
MR. KRESS: So we'll try to wrap it up then tomorrow, and it
will be more of a roundtable discussion.
MR. APOSTOLAKIS: Is NEI going to be here tomorrow?
MR. KRESS: You're welcome to be here. So if there are no
other comments from --
MR. GARRICK: Let me remind the ACNW and the ACNW staff that
our meeting will start in ten minutes.
MR. APOSTOLAKIS: And go on for eight hours.
MR. KRESS: With that, I'm going to recess until tomorrow
morning at 8:30.
[Whereupon, at 5:40 p.m., the meeting was recessed, to be
reconvened at 8:30 a.m., on Friday, January 14, 2000.]
Page Last Reviewed/Updated Tuesday, July 12, 2016