[Federal Register Volume 69, Number 233 (Monday, December 6, 2004)]
[Notices]
[Pages 70475-70480]
From the Federal Register Online via the Government Publishing Office [www.gpo.gov]
[FR Doc No: 04-26688]


-----------------------------------------------------------------------

NUCLEAR REGULATORY COMMISSION


Notice of Availability of Interim Staff Guidance Documents for 
Fuel Cycle Facilities

AGENCY: Nuclear Regulatory Commission.

ACTION: Notice of availability.

-----------------------------------------------------------------------

FOR FURTHER INFORMATION CONTACT: Wilkins Smith, Project manager, 
Technical Support Group, Division of Fuel Cycle Safety and Safeguards, 
Office of Nuclear Material Safety and Safeguards, U.S. Nuclear 
Regulatory Commission, Washington, DC 20005-0001. Telephone: (301) 415-
5788; fax number: (301) 415-5370; e-mail: [email protected].

SUPPLEMENTARY INFORMATION:

I. Introduction

    The Nuclear Regulatory Commission (NRC) plans to issue Interim 
Staff Guidance (ISG) documents for fuel cycle facilities. These ISG 
documents provide clarifying guidance to the NRC staff when reviewing 
either a license application or a license amendment request for a fuel 
cycle facility under 10 CFR part 70. The NRC is soliciting public 
comments on the ISG documents which will be considered in the final 
versions or subsequent revisions.

II. Summary

    The purpose of this notice is to provide the public an opportunity 
to review and comment on a draft Interim Staff Guidance document for 
fuel cycle facilities. Interim Staff Guidance-10 provides guidance to 
NRC staff relative to determining whether the minimum margin of 
subcriticality (MoS) is sufficient to provide an adequate assurance of 
subcriticality for safety to demonstrate compliance with the 
performance requirements of 10 CFR 70.61(d).

III. Further Information

    The document related to this action is available electronically at 
the NRC's Electronic Reading Room at http://www.nrc.gov/reading-rm/adams.html. From this site, you can access the NRC's Agencywide 
Document Access and Management System (ADAMS), which provides text and 
image files of NRC's public documents. The ADAMS ascension number for 
the document related to this notice is ML043290270. If you do not have 
access to ADAMS or if there are problems in accessing the document 
located in ADAMS, contact the NRC Public Document Room (PDR) Reference 
staff at 1-800-397-4209, 301-415-4737, or by e-mail to [email protected].
    This document may also be viewed electronically on the public 
computers located at the NRC's PDR, O 1 F21, One White Flint North, 
11555 Rockville Pike, Rockville, MD 20852. The PDR reproduction 
contractor will copy documents for a fee. Comments and questions should 
be directed to the NRC contact listed above by January 5, 2005. 
Comments received after this date will be considered if it is practical 
to do so, but assurance of consideration cannot be given to comments 
received after this date.

    Dated at Rockville, Maryland, this 24th day of November 2004.

    For the Nuclear Regulatory Commission.
Melanie A. Galloway,
Chief, Technical Support Group, Division of Fuel Cycle Safety and 
Safeguards, Office of Nuclear Material Safety and Safeguards.

Draft--Division of Fuel Cycle Safety and Safeguards Interim Staff 
Guidance--10; Justification for Minimum Margin of Subcriticality for 
Safety Issue

    Technical justification for the selection of the minimum margin of 
subcriticality (MoS) for safety, as required by 10 CFR 70.61(d)

Introduction

    10 CFR 70.61(d) requires, in part, that licensees demonstrate that 
``under normal and credible abnormal conditions, all nuclear processes 
are subcritical, including use of an approved margin of subcriticality 
for safety.'' To demonstrate subcriticality, licensees perform 
validation studies in which critical experiments similar to actual or 
anticipated calculations are chosen and are then used to establish a 
mathematical criterion for subcriticality for all future calculations. 
This criterion is expressed in terms of a limit on the maximum value of 
the calculated keff, which will be referred to in this ISG 
as the upper subcritical limit (USL). The USL includes allowances for 
bias and bias uncertainty as well as an additional margin which will be 
referred to hereafter as the minimum margin of subcriticality (MoS). 
This MoS has been variously referred to within the nuclear industry as 
subcritical margin, arbitrary margin, and administrative margin. The 
term MoS will be used throughout this ISG for consistency, but these 
terms are frequently used interchangeably. This MoS is an allowance for 
any unknown errors in the calculational method that may bias the result 
of calculations, beyond those accounted for explicitly in the 
calculation of the bias and bias uncertainty.
    There is little guidance in the fuel facility Standard Review Plans 
(SRPs) as to what constitutes an acceptable MoS. NUREG-1520, Section 
5.4.3.4.4, states that the MoS should be pre-approved by the NRC and 
that the MoS must ``include adequate allowance for uncertainty in the 
methodology, data, and bias to assure subcriticality.'' However, there 
is little guidance on how to determine the amount of MoS that is 
appropriate. Partly due to the historical lack of guidance, there have 
been significantly different margins of subcriticality approved for 
different fuel cycle facilities over time. In addition, the different 
ways of defining the MoS and calculating keff limits 
significantly compound the potential for confusion. The MoS can have a 
significant effect on facility operations (e.g., storage capacity and 
throughput) and there has therefore been considerable recent interest 
in decreasing the margins of subcriticality below what has been 
accepted historically. These two factors--the lack of guidance and the 
increasing interest in reducing margins of subcriticality--make 
clarification of what constitutes acceptable justification for the MoS 
necessary. In general, consistent with a risk-informed approach to 
regulation, smaller margins of subcriticality require more substantial 
technical justification.
    The purpose of this ISG therefore is to provide guidance on 
determining whether the MoS is sufficient to provide

[[Page 70476]]

an adequate assurance of subcriticality for safety, in accordance with 
10 CFR 70.61(d).

Discussion

    The neutron multiplication factor of a fissile system 
(keff) depends, in general, on many different physical 
variables. The factors that can affect the calculated value of 
keff may be broadly divided into the following categories: 
(1) Geometric form; (2) material composition; and (3) neutron 
distribution. The geometric form and material composition of the system 
determine--together with the underlying nuclear data (e.g., v, X(E), 
and the set of cross section data)--the spatial and energy distribution 
of neutrons in the system (i.e., flux and energy spectrum). An error in 
the nuclear data or in the modeling of these systems can produce an 
error in the calculated value of keff. This difference 
between the calculated and true value of keff is referred to 
as the bias\1\. The bias is defined as the difference between the 
calculated and true values of keff, by the following 
equation: [beta] = kcalc - ktrue
---------------------------------------------------------------------------

    \1\ There are many different ways of computing bias as used in 
calculation of the USL. This may be an average bias, a least-squares 
fitted bias, a bounding bias, etc., as described in the applicant's 
methodology.
---------------------------------------------------------------------------

    The bias of a critical experiment may be known with a high degree 
of confidence because the true (experimental) value is known a priori 
(ktrue [ap] 1). Because both the experimental and the 
calculational uncertainty are known, there is a determinable 
uncertainty associated with the bias. The bias for a calculated system 
other than a critical experiment is not typically known with this same 
high degree of confidence, because ktrue is not typically 
known. The MoS is therefore an allowance for any unknown errors that 
may affect the calculated value of keff, beyond those 
accounted for explicitly in the bias and bias uncertainty. An MoS is 
needed because the critical experiments chosen will, in general, 
exhibit somewhat different geometric forms, material compositions, and 
neutron spectra from those of actual system configurations, and the 
effect of these differences is difficult to quantify. Bias and bias 
uncertainty are estimated by calculating the keff of 
critical experiments with geometric forms, material compositions, and 
neutron spectra similar to those of actual or anticipated calculations. 
However, because of the many factors that can effect the bias, it must 
be recognized that this is only an estimate of the true bias of the 
system; it is not possible to guarantee that all sources of error have 
been accounted for during validation. Thus, use of a smaller MoS 
requires a greater level of assurance that all sources of uncertainty 
and bias have been taken into account and that the bias is known with a 
high degree of accuracy. The MoS should be large compared to known 
uncertainties in the nuclear data and limitations of the methodology 
(e.g., modeling approximations, convergence uncertainties). It should 
be noted that this MoS is only needed when subcritical limits are based 
on the use of calculational methods, including computer and hand 
calculations. The MoS is not needed when subcritical limits are based 
on other methods, such as experiment or published data (e.g., widely 
accepted handbooks or endorsed industry standards).
    Because the nuclear industry has employed widely different 
terminology regarding validation and margin, it is necessary to define 
the following terms as used in this ISG. These definitions are for 
clarity only and are not meant to prescribe any particular terminology.
    Bias: The difference between the calculated and true values of 
keff for a fissile system or set of systems.
    Bias Uncertainty: The calculated uncertainty in the bias as 
determined by a statistical method.
    Margin of subcriticality (MoS): Margin in keff applied 
in addition to bias and bias uncertainty to ensure subcriticality (also 
known as subcritical, arbitrary, or administrative margin). This term 
is shorthand for ``minimum margin of subcriticality''.
    Margin of safety: Margin in one or more system parameters that 
represents the difference between the value of the parameter at which 
it is controlled and the value at which the system becomes critical. 
(This represents an additional margin beyond the MoS.)
    Upper Subcritical Limit: The maximum allowable keff 
value for a system. Generally, the USL is defined by the equation USL = 
1-bias-bias uncertainty-MoS.
    Subcritical Limit: The value of a system parameter at which it is 
controlled to ensure criticality safety, and at which keff 
does not exceed the USL (also known as safety limit).
    Operating Limit: The value of a system parameter at which it is 
administratively controlled to ensure that the system will not exceed 
the subcritical limit.\2\
---------------------------------------------------------------------------

    \2\ Not all licensees have a separate subcritical and operating 
limit. Use of administrative operating limits is optional, because 
the subcritical limit should conservatively take parametric 
tolerances into account.
---------------------------------------------------------------------------

    If the USL is defined as described above, then the MoS represents 
the difference between the average calculated keff 
(including uncertainties) and the USL, thus:
    MoS = (1-bias-bias uncertainty)-USL.
    There are many factors that can affect the code's ability to 
accurately calculate keff and that can thus impact the 
analyst's confidence in the estimation of the bias. Some of these 
factors are described in detail below.

Benchmark Similarity

    Because the bias of calculations is estimated based on critical 
benchmarks with similar geometric form, material composition, and 
neutronic behavior to the systems being evaluated, the degree of 
similarity between benchmarks and actual or anticipated calculations is 
a key consideration in determining the appropriate MoS. The more 
closely the benchmarks represent the characteristics of systems being 
validated, the more confidence exists in the calculated bias and bias 
uncertainty.
    Allowing a comparison of the chosen benchmarks to actual or 
anticipated calculations requires that both the experiments and the 
calculations be described in sufficient detail to permit independent 
verification of results. This may be accomplished by submitting input 
decks for both benchmarks and calculations, or by providing detailed 
drawings, tables, or other such data to the NRC to permit a detailed 
comparison of system parameters.
    In evaluating benchmark similarity, some parameters are obviously 
more significant than others. The parameters that can have the greatest 
effect on the calculated keff of the system are those that 
are most significant. Historically, some parameters have been used as 
trending parameters because these are the parameters that are expected 
to have the greatest effect on the bias. They include the moderator-to-
fuel ratio (e.g., H/U, H/X, v\m\/v\f\), isotopic abundance (e.g., 
\235\U, \239\Pu, or overall Pu-content), and parameters characterizing 
the neutron spectrum (e.g., energy of average lethargy causing fission 
(EALF), or average energy group (AEG)). Other parameters, such as 
material density or overall geometric shape, are generally considered 
to be of less importance. Care should be taken that, when basing 
justification for a reduced MoS on the similarity of benchmarks to 
actual or anticipated calculations, all important system 
characteristics that can affect the bias have been taken into 
consideration. There are several ways to demonstrate that the chosen 
benchmarks are sufficiently similar to actual or anticipated 
calculations:

[[Page 70477]]

    1. NUREG/CR-6698, ``Guide to Validation of Nuclear Criticality 
Safety Calculational Method,'' Table 2.3, contains a set of screening 
criteria for determining benchmark applicability. As is stated in the 
NUREG, these criteria were arrived at by consensus among experienced 
NCS specialists and may be considered conservative. The NRC staff 
considers agreement on all screening criteria to be sufficient 
justification for demonstrating benchmark similarity. However, less 
conservative (i.e., broader) screening ranges may be used if 
appropriately justified.
    2. Use of an analytical method that systematically quantifies the 
degree of similarity between benchmarks and design applications, such 
as Oak Ridge National Laboratory's TSUNAMI code in the SCALE 5 code 
package.
    TSUNAMI calculates a correlation coefficient indicating the degree 
of similarity between each benchmark and calculation in pair-wise 
fashion. The appropriate threshold value of the parameter indicating a 
sufficient degree of similarity is an unresolved issue with the use of 
this method. However, the NRC staff currently considers a correlation 
coefficient ck >= 0.95 to be indicative of a strong degree 
of similarity. Conversely, a correlation coefficient < 0.90 should not 
be used as demonstration of benchmark similarity without significant 
additional justification. These observations are tentative and are 
based on the staff's observation that benchmarks and calculations 
having a correlation of at least 95% also appear to be very similar 
based on a traditional comparison of system parameters. TSUNAMI should 
not be used as a ``black box,'' but may be used to inform the benchmark 
selection process, due to the evolving nature of this tool.
    3. Sensitivity studies may be employed to demonstrate that the 
system keff is highly insensitive to a particular parameter. 
In such cases, a significant error in the parameter will have a small 
effect on the system bias. One example is when the number density of 
certain trace materials can be shown to have a negligible effect on 
keff. Another example is when the presence of a strong 
external absorber has only a slight effect on k\eff\. In both cases, 
such a sensitivity study may be used to justify why agreement with 
regard to a given parameter is not important for demonstrating 
benchmark similarity.
    4. Physical arguments may be used to demonstrate benchmark 
similarity. For example, the fact that oxygen and fluorine are almost 
transparent to thermal neutrons (i.e., cross sections are very low) may 
be used as justification for why the differences in chemical form 
between UO2F2 and UO2 may be ignored.
    A combination of the above methods may also prove helpful in 
demonstrating benchmark similarity. For example, TSUNAMI may be used to 
identify the parameters to which keff is most sensitive, or 
a sensitivity study may be used to confirm TSUNAMI results or justify 
screening ranges. Care should be taken to ensure that all parameters 
which can measurably affect the bias are considered when comparing 
chosen benchmarks to calculations. For example, comparison should not 
be based solely on agreement in the \235\U fission spectrum if \238\U 
or \10\B absorption or \1\H scattering have a significant effect on the 
calculated keff. A method such as TSUNAMI that considers the 
complete set of reactions and nuclides present should be used rather 
than relying on a comparison of only the fission spectra. That all 
important parameters have been included can be determined based on a 
study of the keff sensitivity, as discussed in the next 
section. It is especially important that all materials present in 
calculations that can have more than a negligible effect on the bias 
are included in the chosen benchmarks. In addition, it is necessary 
that if the parameters associated with calculations are outside the 
range of the benchmark data, the effect of extrapolating the bias 
should be taken into account in setting the USL. This should be done by 
making use of trends in the bias. Both the trend and the uncertainty in 
the trend should be extrapolated using an established mathematical 
method.
    Some questions that should be asked in evaluating the chosen 
benchmarks include:
     Are the critical experiments chosen all high-quality 
benchmarks from reliable (e.g., peer-reviewed and widely-accepted) 
sources?
     Are the benchmarks chosen taken from independent sources?
     Do the most important benchmark parameters cover the 
entire range needed for actual or anticipated calculations?
     Is the number of benchmarks sufficient to establish trends 
in the bias across the entire range? (The number depends on the 
specific statistical method employed.)
     Are all important parameters that could affect the bias 
adequately represented in the chosen benchmarks?

System Sensitivity

    Sensitivity of the calculated keff to changes in system 
parameters is a closely related concept to that of similarity. This is 
because those parameters to which keff is most sensitive 
should weigh more heavily in evaluating benchmark similarity. If 
keff is highly sensitive to a given parameter, an error in 
the parameter could be expected to have a significant impact on the 
bias. Conversely, if keff is very insensitive to a given 
parameter, then an error would be expected to have a negligible impact 
on the bias. In the latter case, agreement with regard to that 
parameter is not important to establishing benchmark similarity.
    Two major ways to determine the system's keff 
sensitivity are:
    1. The TSUNAMI code in the SCALE 5 code package can be used to 
calculate the sensitivity coefficients for each nuclide-reaction pair 
present in the problem. TSUNAMI calculates both an integral sensitivity 
coefficient (i.e., summed over all energy groups) and a sensitivity 
profile as a function of energy group. The sensitivity coefficient is 
defined as the fractional change in keff for a 1% change in 
the nuclear cross section. It must be recognized that TSUNAMI only 
evaluates the keff sensitivity to changes in the nuclear 
data, and not to other parameters that could affect the bias and should 
be considered.
    2. Direct sensitivity calculations can also be used to perturb the 
system and gauge the resulting effect on keff. Perturbation 
of the atomic number densities can also be used to confirm the integral 
sensitivity coefficients calculated by TSUNAMI (as when there is doubt 
as to convergence of the adjoint flux).
    The relationship between the keff sensitivity and 
confidence in the bias is the reason that high-enriched uranium fuel 
facilities have historically required a greater MoS than low-enriched 
uranium facilities. High-enriched systems tend to be much more 
sensitive to changes in the underlying system parameters, and in such 
systems, the effect of any errors on the bias would be greatly 
magnified. For this same reason, systems involving weapons-grade 
plutonium would also be more susceptible to undetected errors than low-
assay mixed oxide (i.e., a few percent Pu). The appropriate amount of 
MoS should therefore be commensurate with the sensitivity of the system 
to changes in the underlying parameters.
    Some questions that should be asked in evaluating the 
keff sensitivity include:
     How sensitive is keff to changes in the 
underlying nuclear data (e.g., cross sections)?
     How sensitive is keff to changes in the 
geometric form and material composition?

[[Page 70478]]

     Is the MoS large compared to the expected magnitude of 
changes in keff resulting from errors in the underlying 
system parameters?

Neutron Physics of the System

    Another consideration that may affect the appropriate MoS is the 
extent to which the physical behavior of the system is known. Fissile 
systems which are known to be subcritical with a high degree of 
confidence do not require as much MoS as systems where subcriticality 
is less certain. An example of a system known to be subcritical would 
be a finished fuel assembly. These systems typically can only be made 
critical when highly thermalized, and due to extensive analysis and 
reactor experience, the flooded case is known to be subcritical in 
isolation. In addition, the thermal neutron cross sections for 
materials in finished reactor fuel have been measured with an 
exceptionally high degree of accuracy (as opposed to the unresolved 
resonance region). Other examples may include systems consisting of 
very simple geometry or other idealized situations, in which there is 
strong evidence that the system is subcritical based on comparisons 
with highly similar systems in published references such as handbooks 
or standards. In these cases, the amount of MoS needed may be 
significantly reduced.
    An important factor in determining that the neutron physics of the 
system is well-known is ensuring that the configuration of the system 
is fixed. For example, a finished fuel assembly is subject to tight 
quality assurance checks and has a form that is well-characterized and 
highly stable. A solution or powder process with a complex geometric 
arrangement would be much more susceptible to having its configuration 
change to one whose neutron physics is not well-understood. Experience 
with similar processes may also be credited.
    Some questions that should be asked in evaluating the neutron 
physics of the system include:
     Is the geometric form and material composition of the 
system rigid and unchanging?
     Is the geometric form and material composition of the 
system subject to strict quality assurance?
     Are there other reasons besides criticality calculations 
to conclude that the system will be subcritical (e.g., handbooks, 
standards, reactor fuel studies)?
     How well-known are the cross sections in the energy range 
of interest?

Rigor of Validation Methodology

    Having a high degree of confidence in the estimated bias and bias 
uncertainty requires both that there be a sufficient quantity of well-
behaved benchmarks and that there be a sufficiently rigorous validation 
methodology. If either the data or the methodology is not adequate, a 
high degree of confidence in the results cannot be attained. The 
validation methodology must also be suitable for the data analyzed. For 
example, a statistical methodology relying on the data being normally 
distributed about the mean keff would not be appropriate to 
analyze data that are not normally distributed. A linear regression fit 
to data that has a non-linear bias trend would similarly not be 
appropriate.
    Having a sufficient quantity of well-behaved benchmarks means that: 
(1) There are enough (applicable) benchmarks to make a statistically 
meaningful calculation of the bias and bias uncertainty; (2) the 
benchmarks span the entire range of all important parameters, without 
gaps requiring extrapolation or wide interpolation; and (3) the 
benchmarks do not display any apparent anomalies. Most of the 
statistical methods used rely on the benchmarks being normally 
distributed. To test for normality, there must be a statistically 
significant number of benchmarks (which may vary depending on the test 
employed). If there is insufficient data to verify normality to at 
least the 95% confidence level, then a non-parametric technique should 
be used to analyze the data. In addition, the benchmarks should provide 
a continuum of data across the entire validated range so that any 
variation in the bias as a function of important system parameters may 
be observed. Anomalies that may cast doubt on the results of the 
validation may include the presence of discrete clusters of experiments 
having a lower calculated keff than the set of benchmarks as 
a whole, an excessive fluctuation in keff values (e.g., 
having a X \2\/N [Gt] 1), or discarding an unusually high number of 
benchmarks as outliers (i.e., more than 1-2%).
    Having a sufficiently rigorous validation methodology means having 
a methodology that is appropriate for the number and distribution of 
benchmark experiments, that calculates the bias and bias uncertainty 
using an established statistical methodology, that accounts for any 
trends in the bias, and that accounts for all apparent sources of 
uncertainty in the bias (e.g., the increase in uncertainty due to 
extrapolating the bias beyond the range covered by the benchmark data).
    In addition, confidence that the code's performance is well-
understood means the bias should be relatively small (i.e., bias [lap] 
2%), or else the reason for the bias should be known, and no credit 
must be taken for positive bias. If the absolute value of the bias is 
very large (especially if the reason for the large bias is unknown), 
this may indicate that the calculational method is not very accurate, 
and a larger MoS may be appropriate.
    Some questions that should be asked in evaluating the data and the 
methodology include:
     Is the methodology consistent with the distribution of the 
data (e.g., normal)?
     Are there enough benchmarks to determine the behavior of 
the bias across the entire area of applicability?
     Does the assumed functional form of the bias represent a 
good fit to the benchmark data?
     Are there discrete clusters of benchmarks for which the 
overall bias appears to be non-conservative (especially consisting of 
the most applicable benchmarks)?
     Has additional margin been applied to account for 
extrapolation or wide interpolation?
     Have all apparent bias trends been taken into account?
     Has an excessive number of benchmarks been discarded as 
statistical outliers?
    Performance of an adequate code validation alone is not sufficient 
justification for any specific MoS. The reason for this is that 
determination of the bias and bias uncertainty is separate from 
selection of an appropriate MoS. Therefore, performing an adequate code 
validation is not alone sufficient demonstration that an appropriate 
MoS has been chosen.

Margin in System Parameters

    The MoS is a reflection of the degree of confidence in the results 
of the validation analysis; the MoS is a margin in keff to 
provide a high degree of assurance that fissile systems calculated to 
be subcritical are in fact subcritical. However, there are other types 
of margin that can provide additional assurance of subcriticality; 
these margins are frequently expressed in terms of the system 
parameters rather than keff. It is generally acknowledged 
that the margin to criticality in system parameters (termed the margin 
of safety) is a better indication of the inherent safety of the system 
than margin in keff. In addition to establishing subcritical 
limits on controlled system parameters,

[[Page 70479]]

licensees frequently establish operating limits to ensure that 
subcritical limits are not exceeded. The difference between the 
subcritical limit and the operating limit (if used) of a system 
parameter represents one type of margin that may be credited in 
justifying a lower MoS than would be otherwise acceptable. This 
difference between the subcritical limit and the operating limit should 
not be confused with the MoS. Confusion often arises, however, because 
systems in which keff is highly sensitive to changes in 
process parameters may require both: (1) A large margin between 
subcritical and operating limits, and (2) a large MoS. This is because 
systems in which keff is highly sensitive to changes in 
process parameters are highly sensitive to normal process variations 
and to any potential errors. Both the MoS and the margin between the 
subcritical and operating limits are thus dependent on the 
keff sensitivity of the system.
    In addition to the margin between the subcritical and operating 
limits, there is also usually a significant amount of conservatism in 
the facility's technical practices with regard to modeling. In 
criticality calculations, controlled parameters are typically analyzed 
at their subcritical limits, whereas uncontrolled parameters are 
analyzed at their worst-case credible condition. In addition, 
tolerances must be conservatively taken into account. These technical 
practices generally result in conservatism of at least several percent 
in keff. Examples of this conservatism may include assuming 
optimum concentration in solution processes, neglect of neutron 
absorbers in structural materials, or requiring at least a 1-inch, 
tight-fitting reflector around process equipment. The margin due to 
this conservatism may be credited in justifying a smaller MoS than 
would otherwise be found acceptable. However, in order to take credit 
for this as part of the basis for the MoS, it should be demonstrated 
that the technical practices committed to in the license application 
will result in a predictable and consistent amount of conservatism in 
keff. If this modeling conservatism will not always be 
present, it should not be used as justification for the MoS.
    Some questions that should be asked in evaluating the margin in 
system parameters include:
     How much margin in keff is present due to 
conservatism in the modeling practices?
     Will this margin be present for all normal and credible 
abnormal condition calculations?

Normal vs. Abnormal Conditions

    Historically, several licensees have distinguished between normal 
and abnormal condition keff limits, in that they have a 
higher keff limit for abnormal conditions. Separate limits 
for normal and abnormal condition keff values are 
permissible but are not required.
    There is a certain likelihood associated with the MoS that 
processes calculated to be subcritical will in fact be critical. A 
somewhat higher likelihood is permissible for abnormal than for normal 
condition calculations. This is because the abnormal condition should 
be at least unlikely to occur, in accordance with the double 
contingency principle. That is, achieving the abnormal condition 
requires at least one contingency to have occurred and is likely to be 
promptly corrected upon detection. In addition, there is often 
additional conservatism present in the abnormal condition because 
uncontrolled parameters are analyzed at their worst-case credible 
conditions.
    As stated in NUREG-1718, the fact that abnormal conditions meet the 
standard of being at least unlikely from the standpoint of the double 
contingency principle may be used to justify having a lower MoS than 
would be permissible for normal conditions. In addition, the increased 
risk associated with the less conservative MoS should be commensurate 
with and offset by the unlikelihood of achieving the abnormal 
condition. That is, the likelihood that a process calculated to be 
subcritical will be critical increases when going from a normal to a 
higher abnormal condition keff limit. If the normal 
condition keff limit is acceptable, then the abnormal limit 
will also be acceptable provided this increased likelihood is offset by 
the unlikelihood of going to the abnormal condition because of the 
controls that have been established. If a single keff limit 
is used (i.e., no credit for unlikelihood of the abnormal condition), 
then it must be determined to be acceptable to cover both normal and 
credible abnormal conditions.

Statistical Arguments

    Historically, the argument has been used that the MoS can be 
estimated based on comparing the results of two statistical methods. In 
the USLSTATS code issued with the SCALE code package there are two 
methods for calculating the USL: (1) The Confidence Band with 
Administrative Margin Approach, which calculates USL-1, and (2) the 
Lower Tolerance Band Approach, which calculates USL-2. The MoS is an 
input parameter to the Confidence Band Approach but is not included 
explicitly in the Lower Tolerance Band Approach. Justification that the 
MoS chosen in the Confidence Band Approach is adequate has been based 
on a comparison of USL-1 and USL-2 (i.e., the condition that USL-1, 
including the chosen MoS, is less than USL-2). However, this 
justification is not sufficient.
    The condition that USL-1 < USL-2 is necessary, but not sufficient, 
to show that an adequate MoS has been selected. These methods are two 
different statistical treatments of the data, and a comparison between 
them can only demonstrate whether the MoS is sufficient to bound 
statistical uncertainties included in the Lower Tolerance Band Approach 
but not included in the Confidence Band Approach. There may be other 
statistical or non-statistical errors in the calculation of 
keff that are not handled in the statistical treatments. 
Therefore, the NRC does not consider this an acceptable justification 
for selection of the MoS.

Regulatory Basis

    In addition to complying with paragraphs (b) and (c) of this 
section, the risk of nuclear criticality accidents must be limited by 
assuring that under normal and credible abnormal conditions, all 
nuclear processes are subcritical, including use of an approved margin 
of subcriticality for safety. [10 CFR 70.61(d)]

Technical Review Guidance

    Determination of an adequate MoS is strongly dependent upon the 
specific processes and conditions at the facility being licensed, which 
is largely the reason that different facilities have been licensed with 
different limits. Judgement and experience must be employed in 
evaluating the adequacy of the proposed MoS. Historically, however, an 
MoS of 0.05 in keff has generally been found acceptable for 
a typical low-enriched fuel fabrication facility. This will generally 
be the case provided there is a sufficient quantity of well-behaved 
benchmarks and a sufficiently rigorous validation methodology has been 
employed. For systems involving high-enriched uranium or plutonium, 
additional MoS may be appropriate to account for the increased 
sensitivity of keff to changes in system parameters. There 
is no consistent precedent for such facilities, but the amount of 
increased MoS should be commensurate with the increased keff 
sensitivity of these systems. Therefore, an MoS of 0.05 in 
keff for low-enriched fuel facilities or an MoS of 0.1 for 
high-

[[Page 70480]]

enriched or plutonium fuel facilities must be justified but will 
generally be found acceptable, with the caveats discussed above\3\.
---------------------------------------------------------------------------

    \3\ NUREG-1718, Section 6.4.3.3.4, states that the applicant 
should submit justification for the MoS, but then states that an MoS 
of 0.05 is ``generally considered to be acceptable without 
additional justification when both the bias and its uncertainty are 
determined to be negligible.'' These statements are inconsistent. 
The statement about 0.05 being generally acceptable without 
additional justification is in error and should be removed from the 
next revision to the SRP.
---------------------------------------------------------------------------

    For facility processes involving unusual materials or new process 
conditions, the validation should be reviewed in detail to ensure that 
there are no anomalies associated with unique system characteristics.
    In any case, the MoS should not be reduced below a minimum of 0.02.
    Reducing the MoS below 0.05 for low-enriched processes or 0.1 for 
high-enriched or plutonium processes requires substantial additional 
justification, which may include:
    1. An unusually high degree of similarity between the chosen 
benchmarks and anticipated normal and credible abnormal conditions 
being validated.
    2. Demonstration that the system keff is highly 
insensitive to changes in underlying system parameters, such that the 
worst credible modeling or cross section errors would have a negligible 
effect on the bias.
    3. Demonstration that the system being modeled is known to be 
subcritical with a high degree of confidence. This requires that there 
be other strong evidence in addition to the calculations that the 
system is subcritical (such as comparison with highly similar systems 
in published references such as handbooks or standards).
    4. Demonstration that the validation methodology is exceptionally 
rigorous, so that any potential sources of error have been accounted 
for in calculating the USL.
    5. Demonstration that there is a dependable and consistent amount 
of conservatism in keff due to the conservatism in modeling 
practices.
    In addition, justification of the MoS for abnormal conditions may 
include:
    6. Demonstration that the increased likelihood of a process 
calculated as subcritical being critical is offset by the unlikelihood 
of achieving the abnormal condition.
    This list is not all-inclusive; other technical justification 
demonstrating that there is a high degree of confidence in the 
calculation of keff may be used.

Recommendation

    The guidance in this ISG should supplement the current guidance in 
the NCS chapters of the fuel facility SRPs (NUREG-1520 and -1718). In 
addition, NUREG-1718, Section 6.4.3.3.4, should be revised to remove 
the following sentence: ``A minimum subcritical margin of 0.05 is 
generally considered to be acceptable without additional justification 
when both the bias and its uncertainty are determined to be 
negligible.''

References

NUREG-1520, ``Standard Review Plan for the Review of a License 
Application for a Fuel Cycle Facility''
NUREG-1718, ``Standard Review Plan for the Review of an Application 
for a Mixed Oxide (MOX) Fuel Fabrication Facility''
NUREG/CR-6698, ``Guide for Validation of Nuclear Criticality Safety 
Calculational Methodology''
NUREG/CR-6361, ``Criticality Benchmark Guide for Light-Water-Reactor 
Fuel in Transportation and Storage Packages''

Approved:--------------------------------------------------------------
Date:------------------------------------------------------------------

Director, FCSS
[FR Doc. 04-26688 Filed 12-3-04; 8:45 am]
BILLING CODE 7590-01-P