of 17
8/16/2019 NRC 95-95, Interpretation
1/17
NRC Staff Interpretation of
95/95 Tolerance Limitsin Safety System Setpoint AnalysisNRC Public Meeting with GE-Hitachi
September 28, 2010NRC Headquarters
6003 Executive Boulevard, Rockville, MD 20852
David Rahn, Sr. Electronics Engineer, NRR/DE/EICB
8/16/2019 NRC 95-95, Interpretation
2/17
Disclaimer
• The following slides summarize the collective opinions of
several members of the NRC staff concerning the
interpretation of statistical representations used in theanalysis of instrument channel performance. This
information is being considered for use in establishing
acceptance review criteria pertinent to the review of
proposed instrument setpoint methodologies. The
information herein represents work in progress, and
does not necessarily represent the concurrence of all
cognizant staff members or a final decision in this matter.
2
8/16/2019 NRC 95-95, Interpretation
3/17
Regulatory Guide 1.105
• Regulatory Position C.1 of Rev. 3 (1999) of Reg
Guide 1.105 states:
– “Section 4 of ISA-S67.04-1994 specifies the
methods, but not the criterion, for combining
uncertainties in determining a trip setpoint and its
allowable values. The 95/95 tolerance limit is an
acceptable criterion for uncertainties. That is, thereis a 95% probability that the constructed limits
contain 95% of the population of interest for the
surveillance interval selected.”
3
8/16/2019 NRC 95-95, Interpretation
4/17
Regulatory Guide 1.105 (continued)
• Revision 2 (1986) of Reg. Guide 1.105 stated:
– “Paragraph 4.3 of the standard specifies the methods
for combining uncertainties in determining a tripsetpoint and its allowable values. Typically, the NRC
staff has accepted 95% as a probability limit for
errors. That is, of the observed distribution of values
for a particular error component in the empirical database, 95% of the data points will be bounded by the
value selected. If the data base follows a normal
distribution, this corresponds to an error distribution
approximately equal to a "two sigma" value.”4
8/16/2019 NRC 95-95, Interpretation
5/17
Further Clarification Planned
• The NRC staff has recognized that these definitions
and statements in the previous and current versions
of RG 1.105 could be enhanced with furtherclarification to ensure that all licensees and NRC
reviewers of safety-related instrument setpoints have
a clear, common understanding of the NRC staff’s
concerns. The following slides represent a proposed
way of providing such clarification.
5
8/16/2019 NRC 95-95, Interpretation
6/17
Confidence Interval Estimation• When the magnitude of a particular performance
characteristic, (such as “power supply effect” or “drift”) of
an instrument population is to be estimated, sampling of
that population is performed, and the characteristic ofinterest is measured for each sample. The mean and
standard deviation of the sample are then determined.
• One is usually interested in finding an interval around the
sample mean such that there is a high probability thatthe actual population mean falls inside of this
interval. This interval is called a confidence interval and
the high probability is called the confidence level.
6
8/16/2019 NRC 95-95, Interpretation
7/17
Confidence Interval Estimation• The NRC staff position is that an acceptable confidence level to use
for such estimates of instrument performance characteristics is
±1.96 σ, which is the confidence interval that is large enough to
ensure, with a 95% confidence, that the true population meanactually lies within this interval.
• If the raw data sample size is not sufficiently large, (e.g., there has
not been an appropriate number of tests or measurements of the
characteristic of interest to support statistical test methods) the
interval should be determined through best-estimate means, such asbounding assumptions supported by adequate technical justification,
(e.g., documented historical operating experience) or through use of
appropriate statistical means (e.g., Student’s t-factors)
7
8/16/2019 NRC 95-95, Interpretation
8/17
Tolerance Intervals• Used to specify with a certain degree of confidence
that a specified proportion of a population lies within a
certain interval.
• The “level of confidence” of a tolerance interval is a
measure of the likelihood that the calculated interval of
interest does, in fact, cover at least the specified
proportion of the population of interest.
• The tolerance interval that covers 95% of a population
of interest with 95% confidence is used by the NRC
staff as an acceptance criterion when reviewing
several types of nuclear safety-related analyses.
8
8/16/2019 NRC 95-95, Interpretation
9/17
95% Tolerance Intervals of Interest
0
0.005
0.01
0.015
0.02
0.025
-100.00 -80.00 -60.00 -40.00 -20.00 0.00 20.00 40.00 60.00 80.00 100.00
-1.96 σ +1.96 σ
95% of Probabili ty Distribution2.5% of
Distribution
2.5% ofDistribution
Lower and Lower Probability
Areas
Population Area
of lower interest
(Population area of highest interest)
Highest Probability of Trip
Occurrence
9
Population Area
of lower interest
8/16/2019 NRC 95-95, Interpretation
10/17
Assumptions:• Sufficient data exists to support statistical analyses
• Data representing random error is normally distributed
• Data represents random and independent terms
• Central limit theorem/SRSS methods can be used to
combine like units of standard deviations of random error
• Setpoint errors are disposed symmetrically about the
mean, resulting in the highest probability of a channel trip
being closest to, and symmetric about, the mean
• The highest probabilities of trip occurrence are centered
close to the mean, and lower probabilities are symmetric,
but further away the mean
10
8/16/2019 NRC 95-95, Interpretation
11/17
95% Tolerance Interval• If the endpoints of the interval are being estimated based
on random variables, then the level of confidence
associated with a particular tolerance interval is a
measure of the likelihood that the calculated intervaldoes, in fact, cover at least the specified portion of the
population.
• If the tolerance limits have been based on a statistically
sufficient quantity of sample data, the confidence that theinterval contains 95% of the population of interest
increases.
• The NRC staff uses a confidence level of 95% as an
acceptance criterion for this likelihood. 11
8/16/2019 NRC 95-95, Interpretation
12/17
-0.01
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
-100.00 -80.00 -60.00 -40.00 -20.00 0.00 20.00 40.00 60.00 80.00 100.00
-1.96σ+1.96σ
-1.96σ +1.96σ (for the yellow curve)
(for the red curve)
95% Tolerance Intervals of Interest
for Various Probabil ity Distributions
For each of these curves, 95% of the
population of interest lies between the
interval beginning with -1.96 σ and
ending with + 1.96 σ. The population of
interest is centered about (or, symmetric
with respect to) the mean.
12
8/16/2019 NRC 95-95, Interpretation
13/17
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
1.10
1.20
-12 -7 -2 3 8
-1.96 σ +1.96 σ
+1.645 σ-∞
95% 97.5%
2.5% 5%
Tolerance Interval
of Interest
NEDC-31336P
Tolerance
Interval
Setpoint
5%
2.5%
Cumulative Probabil ity –Interval of
Interest
13
8/16/2019 NRC 95-95, Interpretation
14/17
Cumulative Probability
The NRC staff understanding of the tolerance intervalassociated with the cumulative probability of interest is
that portion of the probability distribution encompassing
the interval from 2.5% of the population to 97.5% of the
population, totaling 95% of the population of interest.
The NRC staff understanding of the 95% tolerance
limits is such that those limits are symmetric with
respect to the mean. In Reg Guide 1.105 Revision 3,
the NRC staff intended that the term “95/95 tolerancelimit” referred to those limits reflecting a 95% confidence
that the limits contain 95% of the population of interest,
which is that centered around the mean, containing the
highest probabilities of occurrence. 14
8/16/2019 NRC 95-95, Interpretation
15/17
Probabil ity of Failure to Trip before
AL is exceeded
• Ignoring instrument channel performance bias terms, an
instrument channel set at 1.96-σ away from the analytical
limit, assuming the 95/95 tolerance interval definition thatincludes the “symmetric about the mean” criterion, has a
probability of failing to trip before the analytical limit is
exceeded of 2.5%.
15
8/16/2019 NRC 95-95, Interpretation
16/17
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
1.10
1.20
-12 -7 -2 3 8
-1.96 σ +1.96 σ
+1.645 σ-∞
95% 97.5%
2.5% 5%
Tolerance Interval
of Interest
NEDC-31336P
Tolerance
Interval
Setpoint
5%
2.5%
AL1 AL2
Instrument Channel Probabil ity of
Failure to Trip before AL: 2.5%
16
Total Loop
Uncertainty
(TLU)
(ignoring
bias terms)
8/16/2019 NRC 95-95, Interpretation
17/17
Other Acceptable Approaches are Possible• Reg Guide 1.105 represents one method of ensuring
that the instrument channel will trip with a 95/95
tolerance limit before the analytical limit is exceeded.
• Other methods are also possible.
• The NRC staff would consider an instrument setpoint
determination method to be adequate where it can be
shown that the instrument channel will trip before the
analytical limit is exceeded, and provided it can beadequately demonstrated that the “95/95 tolerance limit”
is applied to ensure, with 95% confidence, that the
appropriate tolerance interval contains 95% of the
population of interest. 17