NOC-AE-14003103, University of Texas White Paper, Means of Aggregation and NUREG-1829: Geometric and Arithmetic Means, Rev. 3, Enclosure 2 to Attachment 1

University of Texas White Paper, Means of Aggregation and NUREG-1829: Geometric and Arithmetic Means, Rev. 3, Enclosure 2 to Attachment 1

ML14149A437
Person / Time
Site:	South Texas
Issue date:	04/18/2014
From:	Morton D, Pan Y, Tejada J University of Texas at Austin
To:	Office of Nuclear Reactor Regulation
Shared Package
ML14149A439	List: ML14149A434 ML14149A435 NOC-AE-14003103, University of Texas White Paper, Means of Aggregation and NUREG-1829: Geometric and Arithmetic Means, Rev. 3, Enclosure 2 to Attachment 1
References
NOC-AE-14003103, ST133840343, TAC MF2400, TAC MF2401 STP-RIGSI191-ARAI.01, Rev. 3
Download: ML14149A437 (71)

Text

NOC-AE-1 4003103 Attachment 1 Enclosure 2 Enclosure 2 to Attachment 1 University of Texas White Paper supporting resolution of APLAB, Results Interpretation -

Uncertainty Analysis: RAI 2 "Means of Aggregation and NUREG-1829: Geometric and Arithmetic Means", Rev. 3

NOC-AE-14003103 Attachment 1 Enclosure 2 South Texas Project Risk-Informed GSI-191 Evaluation Means of Aggregation and NUREG-1829:

Geometric and Arithmetic Means Document: STP-RIGSI191-ARAI.01 Revision: 3 Date: April 18, 2014 Prepared by:

David Morton, The University of Texas at Austin Ying-An Pan, The University of Texas at Austin Jeremy Tejada, The University of Texas at Austin Reviewed by:

Ernie J. Kee, South Texas Project Zahra Mohaghegh, University of Illinois at Urbana-Champaign Seyed A. Reihani, , University of Illinois at Urbana-Champaign

NOC-AE-14003103 Attachment 1 Enclosure 2 Means of Aggregation and NUREG-1829:

Geometric and Arithmetic Means David Morton, Ying-An Pan, and Jeremy Tejada The University of Texas at Austin Abstract We review methods of combining the probability distributions elicited from multiple experts to obtain a single probability distribution. More specifically, we describe the relative merits of the arithmetic mean (AM) and geometric mean (GM) as ways of performing this aggregation in the context of probabilities associated with rare events. We focus on a study known as NUREG-1829 [9], which includes an expert elicitation of quantiles governing the (annual) frequency of a loss-of-coolant accident (LOCA) in BWRs and PWRs. Examining a set of PWR results from NUREG-1829, we conclude that the GM represents a consistently sensible notion of the middle of the opinions expressed by nine experts. We further conclude that the AM is inappropriate for representing the center of the group's opinion for large effective break sizes. Instead, as the break size grows large a single expert's opinion dominates the combination produced by the AM.

1 Introduction We discuss the arithmetic mean (AM) and geometric mean (GM) as techniques to aggregate prob-ability distributions elicited from multiple experts into a single distribution that represents some notion of group consensus. We specifically study the results obtained by combining the quantiles elicited from experts in Tregoning et al. [9], known as NUREG-1829. This expert elicitation in-volves quantiles (5th, 50th, and 95th percentiles) for exceedance frequencies for a loss-of-coolant accident (LOCA) in BWRs and PWRs. In NUREG-1829, these quantiles are elicited for breaks in the six effective break-size categories given in Table 1 for three time periods: current-day (25 years fleet average), end-of-plant-license (40 years fleet average), and end-of-plant-license-renewal (60 years fleet average). We restrict attention to the PWR fleet, and to illustrate ideas, we restrict attention to the results for the current-day time period.

2 Background Suppose X1, X2, .-. , E R+ represent n data points. There are multiple notions of what constitutes the center of such a data set. The AM and GM represent the center, respectively, via:

AM = -- n1 xi and (la)

GAI = n i 1/n(b 1

NOC-AE-14003103 Attachment 1 Enclosure 2 Table 1: NUREG-1829 LOCA categories for effective break sizes Effective break size (inch) Category 1 category 12 2category 8

3 category 3 7 category 4 14 category 5 31 category 6 The AM and GM satisfy the following inequality:

I) -< n X (2) i=1 i=1 i.e., the GM's value is at most that of the AM. Equality holds in (2) only if x, X2 .... Xn.

Because n(=)1l 1= 0 (n= log(xi))

we can think of the GM as averaging the "exponents" of the data, x1, x,. * *, Xn, rather than directly averaging the x-values. As we will see, if we most naturally visualize the data on a log-scale then there is a strong argument for using the GM in place of the AM, particularly when the data exhibit considerable variability.

We can assign different weights to each data point and generalize the summary measures of equation (1) to:

n wixi and (3a) n f x,, (3b) i=l where the weights wi >_0, i = 1,. . ., n, satisfy Fn=1 w, = 1. If each point is equally weighted, i.e.,

wi = 1/n, i = 1,...,n, then equations (3a) and (3b) reduce to (la) and (lb), respectively. Other widely used notions of the middle of a data set include the median, where the median is defined as follows. Let x(1) < X(2) < ... < x(n) be the n data points reindexed so that they ascend. The 2

NOC-AE-14003103 Attachment 1 Enclosure 2 median defines the middle point by:

x__) or (4a) 2 + ))I where we use (4a) if n is odd and (4b) if n is even.

One important application of the above ideas involves combining probability distributions of multiple experts. Suppose n experts have given a-level quantiles, say, of a LOCA exceedance frequency for a specific break size. If a = 0.5 then these are the median value of each expert's distribution. Denote the quantiles of the n experts by qc,,, qc,2,..., qan. Taking X, = qcj, X2 =

qa,2,. * *, xn = q*,,, we can apply one of the aggregation formulas from (1), (3), or (4) to determine a single a-level quantile that summarizes the views of the experts.

If each expert has a full distribution, i.e., a quantile level for all a G [0, 1], then we can apply an aggregation rule, quantile-by-quantile, to construct a single probability distribution that represents the views of the experts. The equally weighted and unequally weighted arithmetic and geometric means provide a family of ways to combine the distributions of the individual experts.

There is a significant literature on combining expert opinion. See, for example, the discussions in [3, 4, 5, 6] and references therein. It is not our purpose to survey such methods. Nor is it our goal to argue definitively for one method being the universal "right choice." In fact, impossibility theorems-in the spirit of Arrow's seminal work [1]-establish that no rule of aggregation can simul-taneously satisfy certain sets of seemingly compelling properties; see the discussion in French [5].

Two such properties involve: (i) updating the distribution when new information is learned and (ii) marginalizing the distribution by integrating out one component. Updating or marginalizing can first be performed on the distributions of the individual experts, and then the results combined.

Or, the updating or marginalizing can be done on the aggregated, group distribution. The desirable properties are that the results be consistent regardless of which way the updating is done. The GM is consistent under (i) but not (ii) and the opposite result holds for the AM.

Our goal is more empirical in nature and tied to combining the type of rare-event probabilities from multiple experts as investigated in NUREG-1829. Tregoning et al. [9] indicate that the results of NUREG-1829 are sensitive to the method used to combine the individual expert estimates. Our analysis in this report fully agrees that this is the case, particularly as we move from frequencies for small breaks to those for large breaks; i.e., as the associated probabilities shrink and the disparity among expert opinion grows. We do not make sweeping conclusions. Instead, in the context of NUREG-1829, we conclude that the geometric mean represents the middle of the opinions expressed 3

NOC-AE-14003103 Attachment 1 Enclosure 2 by nine experts. And, we conclude that the arithmetic mean is inappropriate for representing the center of the group's opinion, particularly for large effective break sizes.

3 Aggregating Expert Opinion: NUREG-1829 In NUREG-1829, estimates from nine experts were elicited for PWRs for the 5th, 50th, and 95th percentiles for frequencies of the break-size categories 1-6 (see Table 1). Here we focus on the current-day (25 years fleet average) results. The expert elicitation includes piping and nonpiping contributions with multiple subsystems for each. For piping contributions the subsystems include:

Reactor Coolant Piping: Hot Leg; Reactor Coolant Piping: Cold/Crossover Legs; Surge Line; Safety Injection System: Accumulator Lines; Safety Injection System: Direct Volume Injection Lines; Drain Lines; Chemical & Volume Control System; Residual Heat Removal System; Safety Relief Valve Lines; Pressurizer Spray Lines; Reactor Head Lines; and, Instrumentation Lines. For nonpiping contributions the subsystems include: Reactor Pressure Vessel; Pumps; Valves; Pressur-izer; and, Steam Generator. Within each of the nonpiping subsystems, expert opinion was elicited for a number of individual components. For example, for the Steam Generator subsystem, the indi-vidual components are: Tube Rupture; Manway Bolts; Shell; Nozzles; and, Tube Sheet. We point to the Steam Generator subsystem in particular because NUREG-1829 gives summary quantiles for categories 1-6 both with, and without, contributions due to Steam Generator Tube Rupture (SGTR) frequencies. That said, the summary quantiles only differ for categories 1 and 2; i.e., the results with and without SGTR contributions are identical for categories 3-6. To date, STP's pilot effort in a risk-informed resolution of GSI-191 indicates that categories 1 and 2 do not contribute to system failures, and hence the decision to include or exclude SGTR contributions in initiating frequencies appears to have no effect on the analysis. The AM and GM methods for combining the opinions of experts can be applied regardless of whether any interactions between the experts took place. For more on the relative merits associated with various forms of structured interactions among the experts and on various notions of consensus see, e.g., the discussion in [2, 7].

3.1 NUREG-1829

Category 1 NUREG-1829 employs an error-factor adjustment scheme that accounts for possible overconfidence in the opinions of individual experts. We do not detail that scheme here, but roughly speaking, for experts whose uncertainty ranges were relatively small, the error-factor adjustment scheme has the effect of increasing their ranges; i.e., decreasing the 5th percentile and increasing the 95th percentile. Figure 1 shows the 5th, 50th, and 95th percentiles for each of the nine experts after the error-factor adjustment, repeating their labels from NUREG-1829: Experts A, B, C, E, G, H, 4

NOC-AE-14003103 Attachment 1 Enclosure 2 Category 1 Break - Current Day Estimates 12 -- *--AM 11

-- *1-GM Cr' 10 E 9 -0 Expert A

', 8 0 7 6

I~~~~~ i.......... ......... 0 Expert B

-*-Expert C 5 Expert E 0

4 Expert G 3 0- 0 a.

W 2 0 Expert H 1

0 0Expert I 0 0~0 Expert J b

rn.

C:)~C o I"-1 O r qu on

}0:} Io o C) 04Expert L Frequency Figure 1: The 5th, 50th, and 95th percentiles for category 1 exceedance frequencies for nine ex-perts, along with the AM and GM, without SGTR contributions. Note the log-scale of the annual frequency on the x-axis.

I, J, and L. The figure also shows the results of applying the AM and GM formulas from equation (1) to the sets of nine 5th, 50th, and 95th percentiles, respectively. As a first observation, we note that the 5th, 50th, and 95th percentiles of the GM are, respectively, smaller than the 5th, 50th, and 95th percentiles of the AM, as promised by the AM-GM inequality (2).

We seek to understand whether an aggregation rule represents, in some sense, the center or consensus of the group's opinion. To do so we begin by simply counting the number of experts whose values are smaller than and greater than a particular value. For the 50th percentile, calculated according to the AM, there are five experts with smaller medians and four experts with larger medians. We denote this by [5, 6] to indicate that the AM's median falls between those of the 5th and 6th experts, when the expert values are sorted in ascending order for that percentile. Table 2 summarizes the corresponding results for the AM and GM for category 1 under each of the three elicited percentiles.

For each quantile we have nine experts and hence sorting the values, say, for the 50th percentile, the 5th expert is in the middle (at least as measured by the median). The AM and GM values will not (in general) exactly coincide with the value of any single expert. Hence, for the AM and GM the notion of being in the center of the group's opinion is best represented by a result that is either

[4, 5] or [5, 6]. In this sense, the results of Table 2 indicate that both the AM and GM represent the 5

NOC-AE-14003103 Attachment 1 Enclosure 2 Table 2: Characterizing the AM and GM quantiles relative to those of the experts for category 1.

AM GM 5th 50th 95th 5th 50th 95th

[5,6] [5,6] [6,7] [4, 5] [4,5] [4,5]

center of the group's opinions. (The 95th percentile for the AM is slightly larger, falling between the 6th and 7th ascending values of the nine experts.)

We can also compare the AM and GM results by comparing their ratios. We know by the AM-GM inequality (2) that the AM-to-GM ratio exceeds one. Table 3 shows these ratios for the 5th, 50th, and 95th percentiles for category 1.

Table 3: Ratio of AM-to-GNI percentiles for category 1.

5th 50th 95th 2 3 3

3.2 NUREG-1829

Categories 1-6 Examining category 1 in the previous section has allowed us to illustrate ideas. However for GSI-191, results for larger break sizes are of foremost interest. Figure 2 replicates Figure 1 in panel (a),

and includes the other five categories in panels (b)-(f). The plots again account for NUREG-1829's error-factor adjustment scheme. The six plots again capture the 5th, 50th, and 95th percentiles of the nine experts along with the same percentiles for the AM and GM aggregation rules. When examining differences and apparent similarities between the six graphs, we should be careful to note that the log-scale plots of the annual frequencies on the x-axis differ from one plot to the next, descending (very roughly) by an order of magnitude from one category to the next and tending to increase in disparity as the break size grows.

Table 4 extends Table 2 to include the five larger break-size categories. Our observation that the GM represents well the center of the group's distributions for the case of category 1 extends to the other five categories with all of the percentiles falling between the 4th and 5th, or 5th and 6th sorted values, out of nine experts, except for the 5th percentile for category 3. Even for this one apparent aberration, we can see from Figure 2c that the numerical value of the 5th percentile of the GM is relatively close to those of the fourth and fifth sorted 5th percentiles; i.e., the 5th 6

NOC-AE-14003103 Attachment 1 Enclosure 2 percentiles of Experts L and E.

The AM values in Table 4 tell a starkly different story. As the effective break size grows, the quantiles of the AM aggregation scheme tend to become increasingly extreme. In category 6, each of the quantiles of the AM are larger than those of eight of the nine experts, as indicated by [8, 9].

Focusing on category 6, the median of the AM is larger than the 95th percentile of five of the nine experts, and the 5th percentile of the AM is larger than the median of seven of the nine experts.

As can be seen from Figure 2f, Expert A's values are significantly larger than those of the rest of the group. Expert A's opinion dominates the combined opinion when using the AM for category 6, and it is impossible to view the AM as reasonably representing the center of the group's opinion.

Table 4: Characterizing the AM and GM quantiles relative to those of the experts for all categories.

AM GM Category 5th 50th 95th 5th 50th 95th Category 1 [5,6] [5,6] [6,7] [4,5] [4,5] [4,5]

Category 2 [6, 7] [7, 8] [6, 7] [5, 6] [5,6] [5, 6]

Category 3 [8, 9] [8,9] [7,8] [3, 4] [4,5] [4,5]

Category 4 [7,8] [7,81 [6, 7] [5, 6] [4,5] [4,5]

Category 5 [7, 8] [7, 8] [7,8] [4, 5] [5,6] [5,6]

Category 6 [8, 9] [8, 91 [8, 9] [4, 5] [4, 5] [4, 5]

Table 5 extends Table 3 to include results for categories 2-6. We see that the relatively modest ratio of the AM to the GM in category 1 grows large in categories 5 and 6, with the ratios of the medians in category 6 exceeding two orders of magnitude.

Table 5: Ratio of AM-to-GM percentiles for categories 1-6.

Category 5th 50th 95th category 1 2 3 3 category 2 8 8 7 category 3 6 6 5 category 4 5 4 5 category 5 24 22 12 category 6 169 125 64 7

NOC-AE-14003103 Attachment 1 Enclosure 2 Category 1 Break - Current Day Estimates AM S12 9 ExpertA E

o . .. 8p..... .L ............ -

4 Expert B a 7

-*-Expert C 2 10r

-' -Expert E

"-ExpertG

-0 ExpertH Expert I 0

'0- Expert J

• "'-4-Expert L Frequency (a) Category 1 Category 2 Break - Current Day Estimates 12 AM 11 10 .. . . ... ... . .. ............ ° . . . .... . . .....-4*-

4 GM E 9 - -* Expert A 8 --

• Expert B 7 A.,---------

0 6 -*- Expert C

-Expert E 4

0-Expert G 2 ----------- . ... -*- Expert H 1 . ...

-- Expert I 0

0- Expert J

• 4 Expert L Frequency (b) Category 2 Category 3 Break - Current Day Estimates 12 -- *-AM 11 4- GM 10 E 4 9 4 -,Expert A 8 0 Expert B 7

6 _s.

......... .. . . . . . . . . 4 Expert C 5 u Expert E 4

0. Expert G 3

2 I

I................. ...

.... ..........2 .......

=....................................

...... ... 0 Expert H

-*-Expert I 0

b ExpertJ Expert L Frequency (c) Category 3 Figure 2: The 5th, 50th, and 95th percentiles for category 1-6 exceedance frequencies for nine experts, along with the AM and GM, without SGTR contributions. The lack of SGTR contribution only affects categories 1 and 2. Note that the scale on the x-axis changes as the effective break size grows from category 1 to category 6.

8

NOC-AE-14003103 Attachment 1 Enclosure 2 Category 4 Break - Current Day Estimates 12 11 .. . ....... . .... ................. S.... . . . .... .. ....

--- AM

-GM 2 10

• 0 . .. . ..............  : - - O ... "t1 Expert A 48

. .. L0. *......... .. .... Expert B

.7 26 4-Expert C t* 5 Expert E 0Expert G r.. 2

- Expert H 1

0 - Expert I

-* Expert J

-- O-Expert L Frequency (d) Category 4 Category 5 Break - Current Day Estimates 12 .......... - 4 . . .. AM GM 0 -

- w

---

- 41 Expert A I . v. --

, .0 ft.. =* .... Expert B

- Expert C Expert E E ...........

.. Expert G 0

• Expert H

- Expert I 0

Expert J 41Expert L Frequency (e) Category ,5 Category 6 Break - Current Day Estimates 12 -- AM 11 10

......... * ... r.... ..... k

• t = , : 0 -

GM E 9 - Expert A 8  : . . ...... * . ........: .........

........ .." 0 -

Expert B

0. 7 0 Expert C 6 Al--

5 -9 "ExpertE 4 4 -

3 - Expert G 2 4Expert H 1

. ..... . 0 . . . ....... . ...... .. . ... ...... Expert 1 0

Expert J 6 " Expert L Frequency (f) Category 6 Figure 2: (cont.) The 5th, 50th, and 95th percentiles for category 1-6 exceedance freequencies for nine experts, along with the AM and GM, without SGTR contributions. The lack of SGTR contribution only affects categories 1 and 2. Note that the scale on the x-axis changes as the effective break size grows from category 1 to category 6.

9

NOC-AE-14003103 Attachment 1 Enclosure 2

3.3 NUREG-1829

Category 6 and Expert A The intuitive reason Expert A dominates the AM is clear: If we elicit opinions from two experts and one estimates 10-10 and the other 10-6 and we average these with the AM we obtain about 2 10-6; i.e., to order of magnitude the AM is 10-6. If we instead elicit opinions from 10 experts and nine estimate 10-10 and one estimates 10-6 then the AM is on the order of 10-7; i.e., the expert with the largest value dominates, particularly when the values are disparate.

We can quantify the extent to which Expert A's opinion dominates the AM results. Consider the variant of the geometric mean given in equation (3b) in which we assign weights w 1 , 12,..., Wn to each of the n experts. Suppose we assign weight w to one expert (Expert A in the case of category

6) and we weight the remaining experts equally; i.e., with weight (1 - w)/(n - 1). What weight would we need to assign to Expert A in the weighted GM in order to obtain the AM? Assume we seek the weight on the n-th expert. The answer is then given by:

log(AM) - n- 1i og log(x11)

S - -n-- ' log(xi)

In the case of NUREG-1829's category 6, we have n = 9 and we seek the weight on Expert A.

This weight varies slightly for each of the three percentiles, and the values are given in Table 6.

For the median value (50th percentile) we see from the table that we must assign 72.4% weight to Expert A. This means that we assign weight 3.5% to each of the remaining eight experts.

Table 6: Weight required on Expert A in the weighted GM to achieve the AM.

5th 50th 95th 73.6% 72.4% 70.1%

3.4 NUREG-1829's Recommendations on AM versus GM In its Executive Summary, NUREG-1829 reports results for the GM. However, the body of the report also includes results for the AM, and the Executive Sunmmary concludes by saying:

Because alternative aggregation methods can lead to significantly different results, a particular set of L OCA frequency estimates is not generically recommended for all risk-informed applications. The purposes and context of the application must be considered when determining the appropriatenessof any set of elicitation results. ... While the lack of clear application guidance places an additional burden on the users of the study -results, 10

NOC-AE-14003103 Attachment I Enclosure 2 those users are in the best position to judge which study results are most appropriate to consider for their particular applications.

We see the above quote as consistent with the notion of a risk-informed, rather than a risk-based, analysis. Large effective break sizes dominate failure scenarios in STP's risk-informed approach to GSI-191. A natural sensitivity analysis that we can perform would answer the question: What weight must be placed on Expert A, using the ideas in Section 3.3, in order to depart Region III?

Despite the above caveat NUREG-1829 contains in its Executive Summary, the body of NUREG-1829, along with the literature on combining expert opinion, makes a strong case for using GM rather than AM at least when: (i) the elicited probabilities concern rare-event probabilities, (ii) the opinions of the individual experts are disparate, and (iii) we seek a combination rule that represents a reasonable notion of the center of the group's opinion. NUREG-1829 recasts (iii) by saying that validity of the GM aggregation scheme, i.e., the validity of seeking a reasonable notion of the center, assumes no systematic bias:

A basic premise in using an elicitation process is that the panel responses as a whole have no significant systematic bias. While individual responses can be highly uncertain and differ drastically, they do not systematically over- or underestimate the quantities of interest. Many elements of the elicitation procedure are designed to achieve this goal.

NUREG-1829 goes on to say:

A consequence of the assumption of no systematic bias is that the group estimate should be somewhere in the middle of the individual estimates, especially if there are wide differences in the results.

As we have also observed in Section 3, NUREG-1829 notes:

... the AM of individual estimates is often not a good measure of the median group opinion when the individual estimates are widely varying. In this case, the AM is dominated by the one or two largest results and cannot be fairly described as a group estimate.

On reasonable notions of the center of the group's opinions, NUREG-1829 indicates:

There is support for the use of the median or GM in the literature.

In previous NRC applications ... the median was used as a group estimate when the individual estimates varied by several orders of magnitude.

11

NOC-AE-14003103 Attachment I Enclosure 2 NUREG-1829 points to Meyer and Booker [8], who say:

To overcome the influence of extreme values when forming an aggregation estimate, use the median or geometric mean.

NUREG-1829 indicates that in their sensitivity analysis the GM and median produced similar results. This is consistent with the GM results we smnmarize in Table 4 of Section 3.

In general, Von Winterfeldt and Edwards [10] prefer AM-style aggregation, but say:

The only context in which we have any reservations about this conclusion is that of very low probabilities-forsuch extreme numbers, we would prefer averaging log odds to averaging probabilities.

As we discuss in Section 2, averaging logarithms (i.e., averaging exponents) is equivalent to the GM.

We conclude with a summary statement from NUREG-1829, which we see as fully consistent with our analysis in Section 3:

Taking these considerations into account, and noting that a sensitivity study showed that there is little difference between using the median or the GM to aggregate the study results ... the GM was chosen as the most appropriate group estimate which utilizes all the individual estimates.

12

NOC-AE-14003103 Attachment 1 Enclosure 2 References

[1] Arrow, K. J. (1951). Social Choice and Individual Values. Wiley, New York.

[2] Budnitz, R. J., G. Apostolakis, D. M. Boore, L. S. Cluff, K. J. Coppersmith, C. A. Cornell, and P. A. Morris (1998). Use of technical expert panels: Applications to probabilistic seismic hazard analysis. Risk Analysis 18, 463-469.

[3] Clement, R. T. and R. L. Winkler (1999). Combining probability distributions from experts in risk analysis. Risk Analysis 19, 187-203.

[4] Cooke, R. M. (1991). Experts in Uncertainty: Opinion and Subjective Probability in Science.

Oxford University Press, New York.

[5] French, S. (1985). Group consensus probability distributions: A critical survey. In Bayesian Statistics 2, pp. 183-197. North-Holland, Amsterdam.

[6] Genest, C. and J. V. Zidek (1986). Combining probability distributions: A critique and anno-tated bibliography. Statistical Science 1, 114-148.

[7] Mansfield, C. (2004). Peer Review of Expert Elicitation. Memo Prepared for the Environmental Protection Agency, RTI International.

[8] Meyer, M. A. and J. M. Booker (1990). Eliciting and Analyzing Expert Judgment: A Practical Guide, NUREG/CR-5424, U.S. Nuclear Regulatory Commission.

[9] Tregoning, R., L. Abramson, and P. Scott (2008). Estimating Loss-of-Coolant Accident (LOCA)

Frequencies Through the Elicitation Process, NUREG-1829, Volume 1, Nuclear Regulatory Com-mission, Washington, DC.

[10] Von Winterfeldt, D. and W. Edwards (1986). Decision Analysis and Behavioral Research.

Cambridge University Press, New York, NY.

13

NOC-AE-1 4003103 Attachment 2 Attachment 2 Response to ESGB Request for Additional Information:

a. Chemical Effects: RAI 12,13,19

b. Coatings: RAI 2, 3, 4, 5, 7

NOC-AE-1 4003103 Attachment 2 Page 1 of 19 ESGB, Chemical Effects: RAI 12 The CHLE test facility included three parallel vertical head loss loops that were intended to allow multiple bed evaluations with each test. The test results suggest there was potentially some bias in head loss results between the three loops. Please describe any evaluations that were performed and lessons learned. Also, please describe any significant modifications that were made to the facility loop during the course of testing and how those modifications may have affected the results.

STP Response:

In the 30-day CHLE tests conducted prior to the license submittal, no aluminum precipitation was observed, as evidenced by aluminum concentrations that remained below predicted solubility limits and no observed increase in turbidity that would have been indicative of a precipitate in solution. Without direct evidence of aluminum precipitation, it was not possible to relate the head loss characteristics of the 30-day CHLE tests to chemical effects. As described in the LAR Enclosure 4-1, page 9 and in more detail in LAR Enclosure 4-3, Section 5.6.3, safety margin was added to the chemical effects contribution to head loss by applying a bump-up factor to the calculated value of conventional head loss based on known physically relevant parameters of debris bed thickness and sump fluid temperature. The chemical head loss bump-up factor did not directly use head loss data from the CHLE tests. As a result, variability between the three head loss columns has no bearing on the STP license submittal.

Although potential bias between the three head loss loops has no effect on the license submittal, STP recognizes that it is a pilot project and that biases in experimental equipment may affect future licensees. A possible bias was observed in two test series prior to the STP license submittal when the three columns were linked to the tank. The reports of these test series were provided to the NRC Staff in STP Letter NOC-AE-14003075 (1). First, in CHLE-008 (2), preformed WCAP precipitates were introduced into the tank while the tank was connected to the columns. One test was conducted with blender-prepared debris beds in all three columns and a second test was conducted with NEI-prepared debris beds in all three columns. In both cases, the head loss in Column 3 increased 20 to 30 minutes sooner than in Columns 1 and 2. Second, in CHLE-010 (3),

head loss increased on different trends in the three columns after the columns (all containing blender-prepared debris beds) were linked to the tank, with the head loss in Column 3 dramatically higher than the other columns. This difference in head loss occurred despite the fact that no corrosion materials, debris, or precipitation products were present in the tank.

Following the STP license submittal, the contractor, UNM, has explored possible causes for the biases observed in CHLE-008 and CHLE-010 (2,3). A factor that may have contributed to the potential bias was the pipe header configuration that supplied water from the tank to the three columns. The columns were located on a common header, with Column 3 being the last column on the header. The tank system was reconfigured after the STP license submittal so that each column is supplied directly from the tank without the presence of a common header. A repeat of the test in CHLE-008 (2)has not yet been conducted to ascertain whether any change in performance has resulted from the piping modification.

NOC-AE-1 4003103 Attachment 2 Page 2 of 19 It should be noted that other tests were performed in which no systematic bias was observed. The trend in head loss in the MBLOCA test, CHLE-012 ý4) was similar in all three columns, which contained NEI-prepared debris beds. In the LBLOCA test, CHLE-014 (5), also with NEI-prepared debris beds, the head loss in Column 3 was between the values for Columns 1 and 2. The bias has only been observed when blender-prepared debris beds were used. Other tests reported in CHLE-008 (2) demonstrated a high degree of inconsistency and instability in blender-prepared debris beds. The instability of the blender-prepared debris beds causes large increases in head loss from small changes in particulate concentration, regardless of whether it is non-chemical or chemical precipitate. The results indicate that the variability between columns is amplified significantly for the blender-prepared debris beds and not evident for the NEI-prepared debris beds. To minimize this effect, debris beds that are more prototypical of the conditions during a LOCA are being studied at UNM. More prototypical debris beds are sensitive to the presence of precipitates, but are substantially more stable than the blender-prepared debris bed. More prototypical beds consist of fiber prepared with the NEI pressure-washing method supplemented with either paint particles or latent debris and paint particles. Recent tests have demonstrated a high degree of stability, reproducibility between replicate tests in the same column, and reproducibility between similar beds in different columns. Work to characterize the head loss of the beds and document the suitability of these beds for future chemical head loss testing is currently ongoing. Information gained during testing after the license application has not changed the conclusions of the analysis conducted for the license application, and STP is not planning any additional chemical effects testing in support of the license application.

REFERENCES:

1. South Texas Nuclear Operating Company, Letter to the Nuclear Regulatory Commission, NOC-AE-14003075. Feb. 27, 2014. (ML14072A076)

2. University of New Mexico, CHLE-008: Debris Bed Preparationand Formation Test Results, Rev. 4. Feb. 3, 2014. (ML14072A082)

3. University of New Mexico, CHLE-010: CHLE Tank Test Results for Blended and NEI Fiber Beds With Aluminum Addition, Rev. 3. Feb. 10, 2014. (ML14072A083)

4. University of New Mexico, CHLE-012: T1 MBLOCA Test Report, Rev. 4. Feb. 18, 2014. (ML14072A084)

5. University of New Mexico, CHLE-014: T2 LBLOCA Test Report, Rev. 3. Feb. 22, 2014. (ML14072A085)

NOC-AE-1 4003103 Attachment 2 Page 3 of 19 ESGB, Chemical Effects: RAI 13a The aluminum source for CHLE tests was aluminum scaffolding removed from the plant.

The scaffolding was described in the test documents (e.g., CHLE-012) as a non-homogenous sample with unknown constituents from years of use which remained after cleaning. During the NRC staff's visit to observe CHLE testing, the staff observed what appeared to be a grout-like material covering a relatively small portion of a test sample.

The submerged scaffolding samples were taken from the side of the scaffolding and had a different texture and appearance than the samples cut for the vapor space. Analysis of unused scaffolding indicated the presence of aluminum phosphate and aluminum oxide/hydroxide scales. The pre-existing scale may have reduced the aluminum released by corrosion. Since the corrosion of aluminum can have a significant impact on whether chemical precipitates form; (a) Please describe the steps taken to ensure that the surface condition of the scaffolding used in the CHLE tests is representative of the remaining aluminum in the plant.

STP Response:

The aluminum scaffold material used in the CHLE tests was a piece of scaffold pick boards (not the scaffold poles) from the STP plant. The scaffold was typical scaffolding that had been in use at the plant for several years. This piece of scaffold was chosen because it was typical of scaffolding stored in containment. Scaffolding in containment is similar to the used condition of the tested material because scaffold are reused over and over again in the contaminated, high radiation areas to avoid spreading contamination and additional personnel exposure from moving them in and out. Note that a fraction of the surface area of the tested material (the edges) was not oxidized because the pieces had to be cut in order to obtain the correct area exposed, which would be representative of damage to pre-existing scale surfaces on scaffolding in containment. To prevent any change to the surface conditions of the scaffolding, the only preparation in the lab was to clean the sample gently with mild laboratory soap to remove particulate and allow to dry.

NOC-AE-1 4003103 Attachment 2 Page 4 of 19 ESGB, Chemical Effects: RAI 13b (b) Please explain if the corrosion behavior of the different parts of the scaffolding (i.e.,

the part used for submerged samples and the part used for vapor samples) was ever compared by placing them in the exact same test conditions, such as in a bench test.

STP Response:

The corrosion behavior of different parts of the scaffolding was not compared in bench corrosion tests. However, other testing was performed that indicated similar surface conditions, which would be expected to produce similar corrosion characteristics.

Surface conditions were determined using scanning electron microscopy (SEM) with energy dispersive spectrometry (EDS) and X-ray photoelectron spectrometry (XPS). The SEM images in Figure 1 show that the surface of portions of the scaffolding used for submerged and vapor space conditions were qualitatively similar at a microscopic level(1 ). The EDS data shown in Table 1 also show qualitative similarities between the portions of scaffolding, in that the surface elemental composition is dominated by aluminum and oxygen, with small amounts of other elements. The XPS spectra in Figure 2 demonstrates that both portions of scaffolding have surface layers consisting of aluminum phosphate (AIPO 4) and aluminum oxide/hydroxide [AIOOH or AI(OH) 3 , which have overlapping 2p spectra]. In both cases, the surface layers were determined to be about 90 percent AIPO 4 and 10 percent AIOOH/AI(OH) 3 (2). Given the similarities in surface conditions, differences in corrosion behavior are not expected.

Figure 1 - Scanning electron microscope images of scaffolding to be used in (A) submerged conditions and (B) vapor space conditions. These images were collected from pieces of scaffolding that were not used in the corrosion tests and represent the condition as received from STP. (1)

NOC-AE-14003103 Attachment 2 Page 5 of 19 Table I - Elemental composition of scaffolding to be used in (A) submerged conditions and (B) vapor space conditions. This data was collected from pieces of scaffolding that were not used in the corrosion tests and represent the condition as received from STP.

Percent mass on scaffolding to Percent mass on scaffolding to Element be used in submerged conditions be used in vapor space conditions O 37.4 29.1 Na 1.3 0.5 Al 49.8 58.7 Si 5.2 5.1 P 3.4 1.7 Ca 0.4 0.9 S 2.5 Cl -- 0.3 K -- 1.3 Fe -- 1.6 Mg -- 0.6 Zn ..

Bend"nEnrg (6V) Biag E.ne'y (e61 Figure 2 - X-ray photoelectron Al 2p spectra of scaffolding to be used in (A) submerged conditions and (B) vapor space conditions. These data were collected from pieces of scaffolding that were not used in the corrosion tests and represent the condition as received from STP. (2)

NOC-AE-1 4003103 Attachment 2 Page 6 of 19 ESGB, Chemical Effects: RAI 13c (c) Please explain if the scaffolding surface condition was evaluated to determine if a LOCA jet or the thermal transient from a LOCA would cause the oxide to spall, potentially resulting in a greater corrosion rate than was observed during the CHLE testing.

STP Response:

The surface condition of the scaffolding has not been exposed to a LOCA jet to determine if the oxide would spall and result in a greater corrosion rate than observed during the CHLE testing. Some scaffolding is stored in two racks on the containment 19' inside the bioshield wall (per Technical Specification Surveillance procedure OPSP03-XC-0002 and plant general procedure OPGP03-ZM-0028), which is an area where it could be partially exposed to jetting from a hypothesized failure producing a jet that is directed toward the storage racks. The scaffolding is tightly packed into seismically-qualified storage racks and, although the racks are open on the sides, one side is protected by a concrete wall, they are constructed with structural steel I-beams, and closed on the ends with structural steel doors to keep stored material from being released during a seismic event. The tight packing, concrete walls, and structural material all serve as barriers that limit exposure to jets. Finally, energetic jets hypothesized to result from a LOCA are short duration (minutes) compared to corrosion time scales (hours). After the jet dies down, the material is no longer exposed to post-LOCA fluid impingement. In addition, all (100% of the surface area) of the aluminum material is considered exposed to containment spray even though it is tightly packed in the storage racks. This conservative assumption is considered to outweigh any consideration of potential increased corrosion of a limited portion of the aluminum due to removal of an oxide layer by impingement.

REFERENCES:

1. University of New Mexico, CHLE-012: T1 MBLOCA Test Report, Rev. 4. Feb. 18, 2014. (ML14072A084)

2. University of New Mexico, CHLE-014: T2 LBLOCA Test Report, Rev. 3. Feb. 22, 2014. (ML14072A085)

NOC-AE-1 4003103 Attachment 2 Page 7 of 19 ESGB, Chemical Effects: RAI 19 The caption for Figure 5.6.6 (Volume 3) states "Typical sample of sump-strainer head loss histories generated under the assumption of exponential chemical effects factor and artificial head-loss inflation." Please clarify if the "artificial" head loss inflation refers to the NUREGICR-6224, "Parametric Study of the Potential for BWR ECCS Strainer Blockage Due to LOCA Generated Debris," October 1995 (ADAMS Accession No. ML083290498),

correlation multiplied by 5 or some other value.

STP Response:

Yes, the term "artificial" in the caption of Figure 5.6.6 refers to the head loss inflation factor distribution of 5 +/- 1 that is randomly sampled and applied to the NUREG/CR-6224 result for conventional debris head loss.

NOC-AE-1 4003103 Attachment 2 Page 8 of 19 ESGB, Coatings: RAI 2 Table 2.2.16 in Volume 3 provides the quantity of qualified coatings generated within the Zone of Influence (ZOI) for various pipe diameters. The ZOI used to calculate these quantities is not provided. Please provide the ZOI used for both epoxy and inorganic zinc (IOZ) qualified coatings (e.g. epoxy = 4D, IOZ = 10D).

STP Response:

A ZOI was created for each of the four different break sizes in three bounding locations which determine a realistic maximum amount of surface area for the different epoxy, polyamide primer, and IOZ coatings. Due to the different inner diameters of pipes, the 4D ZOI radius is used for the epoxy and inorganic zinc (IOZ) qualified coatings as suggested by WCAP-1 6568-P.

NOC-AE-14003103 Attachment 2 Page 9 of 19 ESGB, Coatings: RAI 3 Section 5.4.5, "Unqualified Coating Debris," in Volume 3 states that the total failure fraction is assumed to be 100% for all unqualified coatings. Given this statement, please describe the significance of the failure fraction analysis provided on pages 11 through 17 in Volume 6.2. Please clarify if the failure fraction analysis used in any manner in the CASA model or if the unqualified coatings are always assumed to have a 100 percent failure fraction.

STP Response:

The failure fraction distributions given in LAR Encl. 5 (Pgs. 12-17) display the results of unqualified coatings failure analysis (LAR Encl. 4-3, Ref. [12]), and show that unqualified coatings (investigated in the STP evaluation) have probable failure fraction ranges. This failure fraction analysis is useful because it shows that failure fractions other than 100%

are probable for unqualified coatings.

However, the failure fraction analysis is not used in the CASA Grande evaluation. All unqualified coatings failure fractions were (assumed) set to 100%.

NOC-AE-1 4003103 Attachment 2 Page 10 of 19 ESGB, Coatings: RAI 4 For unqualified coatings that are not located in the upper containment, the NRC staff's understanding is that 100 percent of the coatings are assumed to fail and are available for transport to the strainer. The staff also understands that 100 percent of the coatings that are calculated to transport to the strainer are assumed to arrive at the strainer uniformly over the first 36 hours4.166667e-4 days <br />0.01 hours <br />5.952381e-5 weeks <br />1.3698e-5 months <br />. Please confirm that the staff's understanding is correct.

Also, please provide details related to the unqualified coating failure assumptions in terms of percentages, timing and quantities that arrive at the sump strainer.

STP Response:

Yes. For unqualified coatings not located in the upper containment, 100% of the coatings are assumed to fail and are assigned a 100% failure fraction in the CASA Grande input deck.

The amount of unqualified coatings (not in upper containment) available for transport is 100%. However, the failed unqualified coatings (not in upper containment) are subject to recirculation transport fractions that designate the fraction (of unqualified coatings) trapped as sediment and the fraction that actively recirculates in the containment pool.

Yes. 100% of the coatings that are calculated to transport to the strainer are assumed to arrive at the strainer. The unqualified coatings quantities calculated to transport to the strainer are the quantities that have already been reduced or multiplied by the failure (100%) and transport fractions (location, recirculation).

CASA Grande intended that the mass of unqualified coatings designated as actively recirculating in the pool be added to the pool at uniform rate over 36 hours4.166667e-4 days <br />0.01 hours <br />5.952381e-5 weeks <br />1.3698e-5 months <br />; however, a code level error (CASA Grande v1.6 Release Notes, Error Report #04) forced the assumption that all actively recirculating materials, including unqualified coatings, arrive in the pool within the first 10 minutes for all scenarios.

Additional information It should be understood that accumulation of suspended debris does not begin until ECCS recirculation begins. A further clarification to the question statement is that uniform debris addition to the pool at a constant rate is not the same as a uniform arrival rate on the strainer. The rate of accumulation on the strainer is driven by total recirculation flow rate, which may change with time.

Unqualified coatings failure fractions/percentages discussed in LAR Encl. 5 are based on analysis of Electric Power Research Institute (EPRI) data (1) performed in LAR Encl. 4-3, Ref. 12 (p. 1-1). The EPRI unqualified coatings study (i., pp. 4-5) reports the percentage of coating detachment from substrates of multiple plant-provided samples for different coating types (e.g. epoxy, alkyd, etc.) subject to design basis preparation (radiation exposure), and autoclave conditions (temperature, pressure, and spray). The EPRI detachment data was used as the basis for the failure fractions described in LAR Encl. 5 (p. 11). CASA Grande did not use these failure fractions for the STP analysis

NOC-AE-1 4003103 Attachment 2 Page 11 of 19 described in LAR Encl. 4 3, in their place, 100% failure was assigned for all unqualified coatings.

The unqualified coatings failure timing was calculated from the EPRI data (pp. 4-5) (1)in LAR Encl. 4-3, Ref. [12] (p. 25). The timing analysis of LAR Encl. 4-3, Ref. [12] (p. 25) is based on visual inspection of filters used in the EPRI unqualified coatings testing (p. 4-3)

(1)* These filters were changed in uneven time increments during the EPRI autoclave testing, and the filters were ranked from 0 to 10 by the degree of discoloration for their respective time increment. The results of the analysis were extrapolated to 30 days with a forced (normalized) cumulative failure of 100% over all time increments. A wash down fraction of 6.0% (LAR Encl. 4-3, Ref. [12], Table 6, p. 30) was assigned in CASA Grande for all unqualified coatings that fail in upper containment to account for sprays being secured within 24 hours2.777778e-4 days <br />0.00667 hours <br />3.968254e-5 weeks <br />9.132e-6 months <br />. The 6.0% washdown fraction is supported by analysis of the EPRI data for failure occurring within 24 hours2.777778e-4 days <br />0.00667 hours <br />3.968254e-5 weeks <br />9.132e-6 months <br />. During STP scenarios where sprays initiate, all sprays are secured close to 6.5 hours5.787037e-5 days <br />0.00139 hours <br />8.267196e-6 weeks <br />1.9025e-6 months <br /> in accord with EOPs with Technical Support Center (TSC) concurrences (LAR Encl. 3-4, Ref. [34]).

The total masses (epoxy, alkyd, etc.) of unqualified coatings entered in CASA Grande were multiplied through their respective transport logic trees of LAR Encl. 4-3 (Figures 5.5.8-5.5.14, pp. 169-172). Percentages listed in the "Fraction of Debris at Sump" column of the transport logic trees (LAR Encl. 4 3, Figures 5.5.8-5.5.14, pp. 169-172) give the fractions available for active circulation in the pool for each branch; the sum of these fractions is given in the bottom right corner of each logic diagram giving the total active circulation fraction ("sum:"). The mass of each unqualified coating multiplied by its respective total active circulation fraction ("sum:") is the amount that was intended to be introduced into the pool over 36 hours4.166667e-4 days <br />0.01 hours <br />5.952381e-5 weeks <br />1.3698e-5 months <br /> (see discussion of Error Report #4 above), but was instead introduced in the first time step inadvertently. All unqualified coatings masses actively recirculating in the pool are calculated to arrive at the strainers according to the solutions of LAR Encl. 4-3, Eqn. 84. Equation 84 models debris accumulation as a cumulative exponential curve defined by the total ECCS flow rate.

Consistent with respective branches in the transport logic diagrams (LAR Encl. 4-3, Figures 5.5.8-5.5.14, Pgs. 169-172), unqualified coatings masses (not in upper containment) added to the pool for active circulation are calculated with Equation A below.

"I$1,,(.4*ve) = Mfcraj

• Ffca - FI.,

• F,,9 -c Equation A where:

MLower(Active)- mass of unqualified coatings not in upper containment available for active circulation Mtotal = Total mass quantity of unqualified coating by type Ffai = Failure fraction of unqualified coating Fl,, = Location transport fraction Frecirc = Transported location fraction (active circulation fraction)

Although all unqualified coatings not in upper containment fail at 100% (Ffaj=l), they are still multiplied by their respective location and recirculation fractions. The location fraction for the unqualified coatings (not in upper containment) Floc assigns the

NOC-AE-1 4003103 Attachment 2 Page 12 of 19 percentage of unqualified coatings that transport to lower containment and the percentage that transports to the reactor cavity (LAR Encl. 4-3, Section 5.5). The active recirculation fraction Frecirc assigns the percentage of failed debris from a location that is actively circulating. Mass totals for each of the unqualified coatings types that are added to the pool for active circulation (Equation A) are shown below in Table A and Table B for lower containment and the reactor cavity, respectively; where transport and failure fractions have been taken from the unqualified coatings debris transport logic diagrams of LAR Encl. 4-3 (Figures 5.5.8-5.5.14, Pgs. 169-172) and total masses taken from LAR Encl. 4-3, Section 2.2.10.

Table A: Actively circulating mass of unqualified coatings found in lower containment Total Failure Location Fraction Transported Mass Actively Coating Type Unqualified Fraction (Lower Recirculation Circulating (Ibm)

C gCoatings MVTotalMass FFfai alFloc Containment) Fraction Frecirc Mower(Active)

M oe(cie Unqualified Alkyd 271 100% 46% 100% 125 (fines)

Unqualified Epoxy 234 100% 2% 100% 5 (fines)

Unqualified Epoxy 709 100% 2% 41% 6 (fine chips)

Unqualified Epoxy 180 100% 2% 0% 0 (small chips)

Unqualified Epoxy 391 100% 2% 0% 0 (large chips)

Unqualified Epoxy 391 100% 2% 100% 8 (curled chips)

Unqualified IOZ 369 100% 17% 100% 63 Coatings (fines)

Table B: Actively circulating mass of unqualified coatings found in the reactor cavity Total Failure Location Transported Mass Actively Coating Type Unqualified Fraction Fraction Recirculation Circulating Coatings Mass FfaiI (Reactor Cavity) Fraction (Ibm)

MTotal Floc Frecirc MLower(Active)

Unqualified Alkyd 271 100% 0% 0% 0 (fines)

Unqualified Epoxy 234 100% 83% 0% 0 (fines) IIII Unqualified Epoxy 709 100% 83% 0% 0 (fine chips)

NOC-AE-1 4003103 Attachment 2 Page 13 of 19 Total Failure Location Transported Mass Actively Coating Type Unqualified Fraction Fraction Recirculation Circulating Coatings Mass Ffail (Reactor Cavity) Fraction (Ibm)

MTotal Floc Frecirc MLower(Active)

Unqualified Epoxy 180 100% 83% 0% 0 (small chips)

Unqualified Epoxy 391 100% 83% 0% 0 (large chips)

Unqualified Epoxy 391 100% 83% 0% 0 (curled chips)

Unqualified IOZ 369 100% 0% 0% 0 Coatings (fines)

REFERENCES:

1. Electric Power Research Institute. Design Basis Accident Testing of Pressurized Water Reactor Unqualified Original Equipment Manufacturer Coatings. Final Report:

September 2005.

NOC-AE-1 4003103 Attachment 2 Page 14 of 19 ESGB, Coatings: RAI 5 Equations 27 and 28 (page 157, Volume 3) refer to F(t) as the "fraction of coatings that fail during a specific time period." Please provide the value of F(t) and describe the analysis performed to arrive at that value for the timeframe during which upper containment spray is active and capable of transporting coatings (the initial 24 hours2.777778e-4 days <br />0.00667 hours <br />3.968254e-5 weeks <br />9.132e-6 months <br />). Please state if F(t) is the same for all unqualified coating types. In addition, please provide the cumulative mass of unqualified coatings that fail in the upper containment in the first 24 hours2.777778e-4 days <br />0.00667 hours <br />3.968254e-5 weeks <br />9.132e-6 months <br /> in the current analysis.

STP Response:

The value of F(t) is 6%. The value (6%) for the 0-24 hour time period was chosen because it represents an over prediction of the time in which all sprays will be secured for LBLOCAs.

Yes. F(t) is the same for all unqualified coating types The cumulative masses of unqualified coatings calculated by LAR Encl. 4-3, Equation 28 are not time dependent and represent the amount of each unqualified coating that would enter the pool over 30 days if sprays were operating through the entire event. This equation and its results are not used in the STP CASA Grande evaluation but values have been provided in Table B below.

The total masses of unqualified coatings that failed in upper containment during the 0-24 hour time period were calculated with Equation 27 and are provided in Table A.

Additional information F(t) was assigned a value of 6% for all unqualified coatings in upper containment, and was applied as a washdown transport fraction in the STP CASA Grande evaluation (LAR Encl. 4-3, Pg. 157). Other values of F(t) are provided in Table 6 of LAR Encl. 4-3, reference 12 (Pg. 30), but they were not applied in the STP CASA Grande evaluation (discussed below).

F(t) values were calculated from EPRI data (i., Pg. 4-5) found in LAR Encl. 4-3, reference 12 (Pg. 25). The unqualified coatings failure timing analysis of LAR Encl. 4-3, reference 12 (Pg. 25) is based on visual inspection of filters used in the EPRI unqualified coatings testing (i., Pg. 4-3). These filters were changed in uneven time increments during the EPRI autoclave testing, and the filters were ranked from 0 to 10 by the degree of discoloration (LAR Encl. 4-3, Ref. [12], Pg. 25) for their respective time increment. The results of the analysis performed in LAR Encl. 4-3, reference 12 (Table 5, Pg. 27) were extrapolated to 30 days with an artificially forced (normalized) cumulative failure of 100%

over all time increments (LAR Encl. 4-3, Ref. [12], Table 6, Pg. 30). The 6% value (0-24 hour time increment) for F(t) was the only time-related failure fraction assigned in CASA Grande for unqualified coatings. The F(t) value (6%) for the 0-24 hour time period was used because it represents an over prediction of the time in which all sprays will be secured for large breaks (LBLOCA) (LAR Encl. Ref. [35]). An F(t) of 6% was only applied to coatings in upper containment. For scenarios where sprays initiate, all sprays are secured close to 6.5 hours5.787037e-5 days <br />0.00139 hours <br />8.267196e-6 weeks <br />1.9025e-6 months <br /> (LAR Encl. 3-4, Ref. [34]).

NOC-AE-1 4003103 Attachment 2 Page 15 of 19 The total mass of each unqualified coating introduced to the pool for active circulation is calculated using the total unqualified coatings masses provided in LAR Encl. 4-3 (Section 2.2.10) and their corresponding debris transport logic diagrams (LAR Encl. 4-3, Figure 5.5.8 - Figure 5.5.14, Pgs. 169 -172). The total mass introduced to the pool for active recirculation includes contributions from the 24 hour2.777778e-4 days <br />0.00667 hours <br />3.968254e-5 weeks <br />9.132e-6 months <br /> 6% washdown fraction representing the time-dependent failure F(t). These total masses were intended to be introduced to the pool at uniform rate over 36 hours4.166667e-4 days <br />0.01 hours <br />5.952381e-5 weeks <br />1.3698e-5 months <br />; however, a code level error (CASA Grande v1.6 Release Notes, Error Report #04) forced addition to the pool of all (actively circulating) materials, including unqualified coatings, at uniform rates during the first 10 minutes. Accumulation of suspended debris does not begin until ECCS recirculation begins.

The total mass of each unqualified coating in upper containment that fails in the first 24 hours2.777778e-4 days <br />0.00667 hours <br />3.968254e-5 weeks <br />9.132e-6 months <br />, and is available for transport, is calculated using Equation 27 below (LAR Encl. 4-3, Pg. 157).

mil(t) = M otalij - Ffoi[,i- F(t) Equation 27 where:

Mij(t) = Mass of unqualified coatings that fail during a specific time period t = Specific time period following the start of the accident Subscript i = Unqualified coating type (epoxy, IOZ, alkyd, or baked enamel)

Subscript j = Coating location (upper containment, lower containment, or reactor cavity)

Mtotau,ij = Total mass of unqualified coatings Ffail " Total failure fraction F(t) = Fraction of coatings that fail during a specific time period Mijcum = Cumulative mass of unqualified coatings that fail The location and unqualified coating type-dependent mass (Mtotai,ij) is calculated as the product Mtotl' 0 = Mtor,,

• F~pp Equation A

where, Mtotai = The total mass of a specific unqualified coating type in containment Fupper - The fraction of a specific unqualified coating in upper containment Substituting Equation A into Equation 27 and evaluating with STP-specific information provides the results shown below in Table A for total failed unqualified coatings mass from the upper containment available to transport to the sump. The right-hand column gives the masses of unqualified coatings (from upper containment) that are in active recirculation during the CASA Grande simulation. These values (Mij of Table A) must be multiplied by their respective recirculation fractions to find the unqualified debris amounts that are actively recirculating in the pool.

NOC-AE-1 4003103 Attachment 2 Page 16 of 19 Table A: Total failed masses from upper containment available to the sump Total Unqualified Upper F(t) Total Failed Coatings Mass Failure Containment Implemented (0-24 hours)

Coating Type (ibm) Fraction Fraction as Washdown Mass in Upper (F. F tFraction Containment (Ibm)

Mtotal Fupper F(t) Mij(t=24 hours)

Unqualified Alkyd (fines) 271 100% 54% 6% 8.8 Unqualified Epoxy (fines) 234 100% 15% 6% 2.1 Unqualified Epoxy (fine chips) 709 100% 15% 6% 6.4 Unqualified Epoxy (small chips) 180 100% 15% 6% 1.6 Unqualified Epoxy 6

(large chips) 391 100% 15%  % 3.5 Unqualified Epoxy (curled chips) 391 100% 15% 6% 3.5 Unqualified IOZ Coatings (fines) 369 100% 83% 6% 18.4 The cumulative mass formulation shown below in Equation 28 (LAR Encl. 4-3, Pg. 157) represents the amount of each unqualified coating that would enter the pool over 30 days if sprays were operating through the entire event. The results of equation 28 are not used in the CASA Grande evaluation.

Mi4.. MAO= o(t) F~il Equation 28 Substituting Equation A into Equation 28 gives results for the cumulative failure (Table B) below.

Table B: Cumulative failed masses from upper containment Total Upper Cumulative Failure (0-Unqualified Failure Containment 24 hrs)

Coating Type Coatings Fraction Fraction Mass in Upper Mass (Ibm) Ffal1 Fr Containment (Ibm)

MtotalFupper Mij,cum Unqualified Alkyd 271 100% 54% 146 (fines)

Unqualified Epoxy 234 100% 15% 35 (fines)

Unqualified Epoxy 709 100% 15% 106 (fine chips)

NOC-AE-1 4003103 Attachment 2 Page 17 of 19 Cumulative Failure (0-Unqualified Failure Containment 24 hrs)

Coating Type Coatings Fraction Continn Mass in Upper Mass (Ibm) Ffal1 Fraction Containment (Ibm)

Mtotal Fuppor Mijcum Unqualified Epoxy 180 100% 15% 27 (small chips)

Unqualified Epoxy 391 100% 15% 59 (large chips)

Unqualified Epoxy 391 100% 15% 59 (curled chips)

Unqualified IOZ 369 100% 83% 306 Coatings (fines)

The cumulative failed masses in the right hand column (Mij,cum) of Table B give the amounts of unqualified coatings that would enter the pool over 30 days if sprays were operating through the entire event. These values (Mij,cum) were not used in the STP CASA Grande evaluation.

REFERENCE:

1. Electric Power Research Institute. Design Basis Accident Testing of Pressurized Water Reactor Unqualified Original Equipment Manufacturer Coatings. Final Report:

September 2005.

NOC-AE-1 4003103 Attachment 2 Page 18 of 19 ESGB, Coatings: RAI 7 Please describe any ongoing containment coating condition assessment program.

Please include the frequency and scope of the inspections, acceptance criteria, and the qualification of personnel who perform containment coatings condition assessment inspections.

STP Response:

The STP coatings are controlled under STP Procedure OPMP06-ZD-0001 "Paints and Coatings". The program includes application requirements, visual examination, and documentation of all concrete and steel-coated surfaces in the containment building to identify any type of coating degradation such as flaking, peeling, blistering, delamination, rusting and mechanical damage. Any areas of degradation are documented and evaluated for severity and determined to be either: repaired during the current outage, repaired in the next available outage, or continued to be monitored and re-evaluated during the next available outage. [Ref: ASTM D 5163-08, para. 10.2 and para. 11.1.2]

The Coating Assessment Inspection includes a visual examination of all accessible Service Level 1 coatings inside containment including the steel containment liner, structural steel, supports, penetrations, uninsulated equipment, and concrete walls and floors receiving epoxy surface systems. This includes areas near sumps associated with the emergency core cooling system. The Coating Assessment Inspection does not include coating of surfaces that are insulated or otherwise enclosed in normal service, and concrete receiving a non-film forming clear sealer coat only. [Ref: ASTM D 5163-08, para. 10.11 The Coating Assessment Inspection of Service Level 1 coatings is conducted at each refueling outage by properly qualified coating inspectors; the time period between outages is approximately 18 months. The Service Level 1 Coatings Assessment Inspection is conducted by STP's Coating Engineer and Coating Planner. The Coating Engineer meets the educational, professional achievements and nuclear coatings experience qualification criteria for qualification as a Nuclear Coatings Specialist in accordance with ASTM D7108-05. The Coating Engineer is the responsible Engineer in charge of the Safety-Related coatings program. The Coatings Planner is a certified NACE Level III Inspector and meets the educational, professional achievements and nuclear coatings experience qualification criteria for qualification as a Nuclear Coatings Specialist in accordance with ASTM D 7108-05. [Ref. ASTM D 5163-08, para. 6.1 and para 9.1].

Prior to the Coating Assessment Inspection, coating inspectors review the two previous coating assessment reports. From the previous two coating assessment reports, areas identified as being monitored and re-evaluated in the next available outage are noted and added to the location maps for the current Coatings Assessment Inspection. [Ref:

ASTM D 5163-08, para. 7.2]

The coating inspectors bring into the Reactor Containment Building (RCB) the proper instruments and equipment needed for inspection, including, but not limited to: location maps, flashlights, marker pen, measuring tape, feeler gauge, binoculars and camera.

The location maps dividing the reactor containment building into twenty-four (24) identifiable floor plans are labeled with pertinent elevation, azimuth references, structural

NOC-AE-1 4003103 Attachment 2 Page 19 of 19 features and components. All areas of degraded coatings identified during the Coatings Assessment Inspection are recorded on the location maps. All areas that cannot be inspected during the Coatings Assessment Inspection and the specific reason why the inspection cannot be conducted are identified on the location maps. [Ref: ASTM D 5163-08, para. 7.2, para. 10.1.3 and para. 10.5]

Physical tests are performed on an as-need basis as determined by the coatings inspectors. Blistering of all sizes, flaking, peeling, and delamination are considered rejectable conditions. The source and extent of any rusting is evaluated during the visual inspection by the coatings inspectors. Cracks over 30 mils are evaluated by Engineering and all cracks less than 30 mils are a rejectable condition and documented in accordance with PGP03ZX0002, Condition Reporting Process. [Ref: ASTM D 5163-08, para. 10.2]

The Coatings Assessment Report has been evaluated and approved by the coatings inspectors who collaborate in the evaluation of degraded coatings and determination of recommendations. The Coatings Engineer prepares the Coatings Assessment Report and the Certified NACE Level III Coatings Planner prepares work orders for the repair of degraded coatings in accordance with PGP03ZX0002, Condition Reporting Process.

[Ref: ASTM D 5163-08, para. 11.1]

NOC-AE-1 4003103 Attachment 3 Attachment 3 Response to SRXB Request for Additional Information: RAI 5, 6, 7, 8, 9

NOC-AE-1 4003103 Attachment 3 Page 1 of 9 SRXB: RAI 5a Table 2.2.1 in Volume 3 provides results for sump switchover time based on the break size during a loss of coolant accident (LOCA). The application states that "the timing for switchover to recirculation is dependent on the volume of water in the RWST and the total ECCS and CSS flow rate."

(a) Please provide the assumptions used for the volume of water in the RWST for the results in Table 2.2.1. Please provide justification for use of these assumptions.

STP Response:

The results shown in Table 2.2.1 were obtained from the simulations of LOCA scenarios of different break sizes performed with the RELAP5-3D and MELCOR input models. A detailed description of the simulation conditions is reported in [1]. The volume of the water in the RWST defined in the RELAP5-3D input model was defined in accordance with the plant emergency operation procedures OPOP05-EO-EO10 step 22 [2] which instruct the operator to initiate the sump switchover procedure when RWST level is less than 75,000 gallons, which is when the low-low level alarm is actuated. Table 1 lists the RWST volumes of water at different alarm levels.

Table 1. RWST Volumes

'Level 3

I, Volume High Alarm ('Nominal) 528,000 Lowý,vAlarm (Nominal) 473,000 Low-Low Alarm (Nominal) 75,000 Empty Volume (Nominal) 32,000 Max tUsabhle Volm 496,000 Min Usable VOIl1me _ 441,000 Avea- UabkýVoluime "! 468,500 Usable Vohinie LOA) 456,735 INectioln Vol9!1me (J-(-CA) 413,735 The Injection was calculated as follow:

Max. (Min.) Usable Volume = High-Alarm (Low Alarm) Volume - Empty Volume Average Usable Volume = (Max Usable Volume + Min Usable Volume)/2 Injection Volume = Average Usable Volume - (Low-Low Alarm Volume - Empty Volume)

The Injection Volume defined in the RELAP5-3D input model is equal to 413,735 US gal.

NOC-AE-1 4003103 Attachment 3 Page 2 of 9 SRXB: RAI 5b (b) Please provide the ECCS flow rate and CSS flow rate for each break size in Table 2.2.1.

STP Response:

Table 2 shows the ECCS flow rate and the CSS flow rate for the break sizes reported on Table 2.2.1.

Table 2. Flow rates and sump switchover time Break Size 1.5" 2" 4" 6" 8" 15" DEG Sump Switchover Time 5.6h 1.3h 55.9m 44.2m 37.8m 31.2m 29.5m Total SI Flow Rate (gpm) 1230.7 2075.8 4119.6 7950.7 10285.4 11779.7 11988.2 Total CS Flow Rate (gpm) 0.01 5200.0 5200.0 5200.0 5200.0 5200.0 5200.0 NOTE 1: Simulation results showed that the containment sprays did not actuate for a 1.5" break.

The containment sprays volumetric flow rate imposed in the MELCOR input model was calculated under the following assumptions:

" The flow rate for each CS pump was imposed to be equal to 2600 gpm (corresponding to the maximum single train flow rate)

" One of the three containment spray pumps manually secured at the beginning of the transient [2].

NOC-AE-1 4003103 Attachment 3 Page 3 of 9 SRXB: RAI 5c (c) Please explain how sump switchover time is calculated based on the responses to

a. and b. above.

STP Response:

The sump switchover time was calculated via a control variable defined in the RELAP5-3D input file as the time required to deplete the RWST injection volume of 413,735 US gal. Details of the control variable logic adopted are listed below:

1) A control variable was defined to calculate the total ECCS flow rate as the sum of the flow rate of all the active safety injection pumps (calculated by the RELAP5-3D simulation) and containment spray pumps (calculated by the MELCOR simulation).

2) An "integral control variable" was defined in the RELAP5-3D input deck to calculate the total volume of water withdrawn from the RWST, as the time integral of the total flow rate described in 1).

t =t total injected 12olume = f (total flow rate)dt'

3) The sump switchover time was calculated as the time t at which the total injected volume calculated in 2) reaches the RWST injection volume.

NOC-AE-14003103 Attachment 3 Page 4 of 9 SRXB: RAI 6 Section 2.2.1 in Volume 3 describes the termination criteria for containment sprays. One of the criteria to terminate containment sprays is that containment pressure has dropped below 6.5 psig. Please provide plots for containment pressure versus time for a range of break sizes to verify pressure drops below the termination criteria of 6.5 pounds per square inch gauge (psig) before 6.5 hours5.787037e-5 days <br />0.00139 hours <br />8.267196e-6 weeks <br />1.9025e-6 months <br />.

STP Response:

Figure 1 shows the containment pressure vs. time estimated using MELCOR for the range of the range of break sizes reported in Table 2.2.1, under the nominal operating conditions [1]:

" Number of operating SI trains = 3

" Number of Containment Sprays = 2 (one pump secured by manual procedure

[2]

" Number of fan coolers = 6

• CCW Temperature = 85.84 OF The spray actuation set point (9.5 psig) and the sprays termination set point (6.5 psig) are also plotted in the figure.

50 45 -

-2"

-4" 140 "

i - 15" 35 0.

9L ...... Termination Set Point = 21.2 ls (6.5 pill V-- -- CS Actuation Set Point =242 psla (9.5 p&WS E30 25 2000 1- , -- - - - -

01 0 500 100 1500 2000 2500 3000 3500 4000 4500 5000 Time [s]

Figure 1. Containment Pressure

NOC-AE-1 4003103 Attachment 3 Page 5 of 9 SRXB: RAI 7a Switchover to hot leg injection is started 5.5 hours5.787037e-5 days <br />0.00139 hours <br />8.267196e-6 weeks <br />1.9025e-6 months <br /> after the beginning of the LOCA event and is assumed to be completed between 5.75 and 6 hours6.944444e-5 days <br />0.00167 hours <br />9.920635e-6 weeks <br />2.283e-6 months <br />.

(a) Please summarize the Emergency Operator Procedures (EOPs) that direct the operators to take this action and an associated timeline of the operator key actions if this event were to occur.

STP Response:

The instructions for transferring the Safety Injection System to hot leg recirculation are provided in the Emergency Operator Procedure OPOP05-EO-ES14 "Transfer to Hot Leg Recirculation". This procedure is entered from the OPOP05-EO-EO10 "Loss of Reactor or Secondary Coolant" Step 28, when 5.5 hours5.787037e-5 days <br />0.00139 hours <br />8.267196e-6 weeks <br />1.9025e-6 months <br /> have elapsed.

The actions followed by the operator during this procedure are summarized in the table below.

NOC-AE-1 4003103 Attachment 3 Page 6 of 9 Switchover to hot leg: Summary of the procedure Step Summary Comment lime required Transfer to hot leg recirculation procedure is entered from OPOPOS-EO-EO1O, "LOSS OF REACTOR OR SECONDARY COOLANT, 0 Step 28, when 5.5 hours5.787037e-5 days <br />0.00139 hours <br />8.267196e-6 weeks <br />1.9025e-6 months <br /> have elapsed ENERGIZE selected SI train HHSI The HHSI hot leg injection valves have power In Control Room hot leg injection valve lockout switches in the Control Room, (no time) allowing power to be restored via handswitches in the control room.

OPEN HHSI hot leg injection valve Select a train to switch to Hot Leg injection, line up and switch. Check that flow is CLOSE HHSI cold leg injection valve established after the transfer occurs.

DEENERGIZE selected SI train HHSI hot leg injection valve Normally de-energized valves {Cold Leg) Two trains swapped normally (swap two trains).

Close the selected LHSI Cold Leg 1.5 minutes travel time to A Train switchgear 2.25 minutes (two IBreake~s: room breakers plus LHSI PUMP A DISCH TO LOOP 1 travel time to COLD LEG switchgear)

OR LHSI PUMP B DISCH TO LOOP 2 1.75 minutes (two COLD LEG room.

1 minute travel time to BTrain switchgear breakers plus OR travel time to switchgear)

LHSI PUMP C DISCH TO LOOP 3 2.25 minutes (two COLD LEG 1.5 minutes travel time to C Train switchgear breakers plus room. travel time to switchgear)

ENERGIZE and OPEN selected LHSI The LHSI hot leg injection valves have power In Control Room hot leg injection valve lockout switches in the Control Room; which (no time) allows power to be restored via handswitches CLOSE selected LHSI cold leg in the control room.

injection valve Pick a train to switch to Hot Leg injection, line up and switch. Check that flow is established after the transfer occurs. Repeat for second train, if available.

NOC-AE-1 4003103 Attachment 3 Page 7 of 9 SRXB: RAI 7b (b) Please provide a justification that demonstrates the 15-30 minute response time is achievable during switchover to hot leg injection training scenarios. In the justification, please include the results of simulator runs and training logs.

STP Response:

The table in the response to RAI 7a above shows the estimated time required to perform the manual operations to complete the hot leg switchover procedure [3]. The time required to complete each operation described in the table were taken from the general operator action transit times and equipment manipulation times from EOPT-05.01 [4]

measured for Emergency Operator Procedures local step timing. These actions are not timed in the'simulator.

As specified in the STP plant emergency procedure [3], only two of the three SI trains are switched to hot leg recirculation. Under these conditions, the total field time required to transfer the SI trains (two) to hot leg recirculation is 4.5 min. Note that this is the local operator field time and some additional time over the field time is required to complete the other steps that are performed in the control room. The additional time required would be to manipulate the control hand switches and perform the step execution (three-way communication and verification, etc.) However, since the other manipulations are in the control room on the same control panel, 15 minutes would be adequate time to complete the evolution.

NOC-AE-1 4003103 Attachment 3 Page 8 of 9 SRXB: RAI 8 Table 2.2.14 (Volume 3) shows safety injection (SI) flow rates for nominal operating conditions. Please justify the use of nominal conditions versus the use of limiting conditions when analyzing LOCA.

STP Response:

The values of the total Safety Injection (SI) volumetric flow rates shown in table 2.2.14 represent the volumetric flow rate at the sump switchover time for different break sizes, calculated with the thermal-hydraulic simulations, assuming that all the SI pumps (HHSI, LHSI, three SI trains) are available during the simulation. All SI pumps operating is defined as the nominal operating condition. For any other given scenario, the volumetric flow rate for individual SI pumps was estimated based on the ratio of the maximum pump capacities, using Equations 4 and 5 (Volume 3, page 53 of 248).

Nominal SI operating conditions are a reasonable surrogate for all plant-state configurations because: (1) nominal conditions are expected for almost 95% of all LOCA accident scenarios, and (2) plant configurations that include failed SI pumps draw water to the sump at a far lower rate than the small increase experienced from relaxation of backpressure on the RCS afforded by the loss of one or more SI pumps.

NOC-AE-1 4003103 Attachment 3 Page 9 of 9 SRXB: RAI 9 Please describe the terms total sump flow, total SI flow, and ECCS flow. Please include if high head safety injection (HHSI), low head safety injection (LHSI), or CSS is a part of each definition.

STP Response:

" The Total SI Flow Rate is the flow rate forced by the HHSI and LHSI of the three SI trains.

Total SI Flow Rate = (HHSI flow rate + LHSI flow rate).

" The Total ECCS Flow Rate is the sum of the Total SI flow rate and the Containment Sprays (CS) Flow Rate.

Total ECCS Flow Rate = HHSI flow rate + LHSI flow rate + CS flow rate

" The Total Sump Flow Rate is the total flow through the sump strainers. The value of the total sump flow rate at any time is equal to the Total ECCS Flow Rate:

Total Sump Flow Rate = Total ECCS Flow Rate

References:

[1]. Texas A&M University, "Sump Temperature Sensitivity Analysis" Revision 2.0. January 2013

[2]. South Texas Project Electric Generation Station, "Loss of Reactor or Secondary Coolant". Department Procedure Safety Related (Q), OPOP05-EO-EO10 Rev.20 (2011)

[3]. South Texas Project Electric Generation Station, "Transfer to Hot Leg Recirculation" Department Procedure Safety Related (Q), OPOP05-EO-ES14 Rev.7 (2008).

[4]. South Texas Project Electric Generation Station, "EOP Local Action. General Operator Action Transit Times" EOPT-05.01 Rev.5.

NOC-AE-1 4003103 Attachment 4 Attachment 4 Response to SSIB Request for Additional Information:

a. ZOI: RAI 1

b. Debris Characteristics: RAI 2 C. Transport: RAI 5, 9, 11, 13

d. NPSH and Degasification: RAI 29

NOC-AE-1 4003103 Attachment 4 Page 1 of 19 SSIB, ZOI: RAI 1 Volume 3, Table 5.3.1 lists Double Ended Guillotine Break (DEGB) equivalent diameters and Table 5.3.2 lists computer aided design (CAD) DEGB values. These values are calculated by doubling the single sided break area and then calculating an equivalent pipe diameter. Approved ZOls are based on doubling the volume of single sided break jets (calculated by the American National Standards Institute (ANSI) model) and calculating a radius for a sphere based on that volume. Please describe how the CAD DEGB values calculated by Equation 22 of Volume 3 are used.

STP Response:

The DEGB values computed by Equation 22 of LAR Encl. 4-3 (Page 125, Section 5.3.1) are extraneous information and were not used in the analysis. Table 5.3.1 (LAR Encl. 4-3, Pg. 126) includes data representing the nominal pipe size, actual pipe size and an unused extrapolation to a DEGB size defined by Equation 22 for each weld category.

These unused CAD calculated DEGB sizes were also apparent in Table 5.3.2 (LAR Encl. 4-3, Pg. 128). The pipe schedule and outer diameter define actual pipe size, synonymous to its inner diameter within LAR Encl. 4-3. CASA Grande analysis utilizes actual pipe size for all calculations, including ZOI determination.

An action to remove the unused information has been entered in the STP corrective action program to track correction for future submittals.

NOC-AE-1 4003103 Attachment 4 Page 2 of 19 SSIB, Debris Characteristics: RAI 2 Please provide the size distributions for Nukon and Thermal Wrap debris created by the postulated LOCA jet (percentage of each size category created). Please provide the methodology used, including the bases, to determine the size distributions. Please provide information regarding whether the distribution is a simple percentage of all generated fibrous debris or based on the distance of the insulation from the break (Volume 3, Sections 2.2.15, "Insulation Debris Size Distribution," 4.2, "Structures Information Process Flow," and 5.4.2, "Insulation Debris Size Distribution Model").

STP Response:

The size distributions for Nukon and Thermal Wrap debris created by the postulated LOCA jet (% of each size category created) are as listed below. These values are contained in reference 46, Table 4.1 (LAR Encl. 4-3).

Nukon and Thermal Wrap are analyzed as low density fiberglass (LDFG) and have the same debris size distribution and ZOls.

The methodology NEI 04-07 Vol. 2, reference 45, was used to determine the size distributions and is based on air-jet impingement tests (AJIT).

The distribution is based on the distance of the insulation from the break and the corresponding impact pressure witnessed at the target.

NOC-AE-1 4003103 Attachment 4 Page 3 of 19 SSIB, Transport: RAI 5 Please explain assumption 6.h.v of Volume 3 (page 79). Please describe how the number of strainers in service affects pool fill transport. It appears that the pool fill transport phase would be largely completed prior to strainers being placed in service.

STP Response:

Assumption 6.h.V states that all three strainers are active during pool fill transport. Each sump, regardless of service state, will collect a quantity of fines during pool fill as listed in Table 5.5.3.

Table 5.5.3 - Pool fill transport fractions used in CASA Grande Debris Type Pool Fill Transport Fractions Each Sump Inactive Cavities Fines 2% 5%

Small LDFG 0% 0%

Large LDFG 0% 0%

NOC-AE-1 4003103 Attachment 4 Page 4 of 19 SSIB, Transport: RAI 9 It was not clear to the NRC staff if the fibrous debris eroded from large and small pieces of debris were added to the fine transport term. Please clarify that the eroded term was added to the fine source term and is not added as the size category from which they were eroded. For example, in Figure 5.5.3 of Volume 3, the transported debris should be 1.8 percent fines and 35.6 percent small while the total shows 37.4 percent transported (35.6 + 1.8). Please state if the evaluation includes the 1.8 percent as fine debris.

STP Response:

Yes, the evaluation includes the 1.8% as fine debris. All eroded fiber (regardless of origin) and all fines and small fibers that have not settled are treated identically as suspendable debris. Figure 5.5.3 of LAR Encl. 4-3 shows that 37.4% of small pieces generated in the ZOI will be introduced to the pool as suspended material that eventually reaches the strainer.

Debris transport equations and the debris transport logic diagrams provided in the LAR Encl. 4-3, Section 5.5.7 describes the total amount of suspendable debris that is introduced to the pool from each low density fiberglass (LDFG) size category. All LDFG that arrives in the pool is either (1) treated as suspended for eventual transport to the strainer, (2) sequestered in inactive cavities, (3) applied directly to the strainers during pool fill, or (4) settled on the floor.

Erosion of large LDFG pieces contributes transportable debris to the pool. Erosion of settled small LDFG pieces into fines also contributes transportable debris to the pool.

Once introduced to the pool, the size category of origin does not affect total transport because no further credit is applied for settling.

NOC-AE-1 4003103 Attachment 4 Page 5 of 19 SSIB, Transport: RAI a1 a

Table 5.5.5 of Volume 3 lists the recirculation debris time, x(t) and states that it is described in Section 5.8 of Volume 3. The NRC staff was not able to locate a description of this variable. Please provide a definition of the debris recirculation time. The staff found the term x(t) described in STP's initial submittal dared June 19, 2013 (page 74 of 174 of the attachment), which has been superseded in its entirety. Please provide a response to the following questions, which are partially based on the information provided in June 19, 2013, submittal.

(a) Please provide a description of the recirculation debris time as implemented in submittal dated November 13, 2013.

STP Response:

The term x(t) describes the time-dependent mass of debris in the pool with an exponential decay function. As written, the term x(t) (LAR. Encl. 4-3, Equation 41, to the June 19, 2013 application ML131750250) should be disregarded.

It was taken out of LAR Encl. 4-3 (current licensing application) because it was not implemented in CASA Grande in this form. CASA Grande tracks each individual debris type with solutions to Equation 84 (LAR Encl. 4-3, Pg. 209), which describes the differential rate of change of debris mass suspended in the pool. The debris recirculation time is accurately defined by the ratio of pool volume (M3) to total volumetric flow rate (m 3/s), which can change as a function of time if pumps are turned off.

The x(t) discrepancy has been added to the STP corrective action program to track correction for future submittals.

NOC-AE-1 4003103 Attachment 4 Page 6 of 19 SSIB, Transport: RAI 11b (b) Please describe the types and sizes of debris to which the recirculation debris time applies.

STP Response:

The term x(t) was not implemented into CASA Grande in the form discussed in LAR Encl. 4-3 (June 19, 2013 application ML131750250, Equation 41). The debris recirculation time is accurately defined by the ratio of pool volume (M3) to total volumetric flow rate (m3/s).

Equation 84 of LAR Encl. 4-3 (current application, Pg. 209) tracks all particulate and fibrous debris types that can be suspended in the pool assuming a homogenously mixed pool. This includes all coatings (Qualified and Unqualified), all other particulates, and all fibrous insulation debris types inside of containment having sizes classified as fines or small pieces. An erosion fraction is applied to large pieces to generate additional suspended fiber. The sizes of all debris types evaluated are listed in Table 2.2.21 of LAR Encl. 4-3 (Pg. 57). The same debris recirculation time applies to all suspended debris.

Equation 84 of LAR Encl. 4-3 (current application, Pg. 209), only applies to those quantities that arrive on the strainer during recirculation. Those materials that are initially transported to the strainer (early arrival) as a result of pool fill and sheeting flow are not dependent on the ratio of pool volume (V) to total recirculation pump flow (Q), but rather, are placed directly on the active strainers at initiation of the LOCA. Their mass however is accounted for and tracked in the total inventory during the scenario.

NOC-AE-1 4003103 Attachment 4 Page 7 of 19 SSIB, Transport: RAI 11c (c) Please state if the equation assumes homogenous mixing in the pool. If so, please explain if this is a valid assumption for all debris types.

STP Response:

Equation 84 of LAR Encl. 4-3 (current application, Pg. 209) used to track pool debris inventory assumes homogeneous mixing in the pool of suspended debris.

This assumption is valid for all suspended debris types for the following reasons:

1. For all breaks, the initial high floor velocities from sheeting flow caused by the pipe break and containment sprays are expected to scatter debris with no preferential direction throughout containment.

2. Fine debris will be further mixed after recirculation because of multidirectional velocity vectors and turbulent kinetic energy.

As an added conservatism, solutions to the pool transport equation (LAR Encl. 4-3, Section 5.8, Equation 84) do not take credit for settling of debris in the pool. This means that all suspended debris is recirculated through the ECCS back into the pool until 100%

is eventually trapped at the strainers or on the reactor core. CFD results for steady state, isothermal recirculation are shown below.

CFD representations of flow vectors for a near-sump and far-sump large LOCA breaks are described in the figures. In the near-sump break simulation (Case 1) a break on the Loop C Hot Leg was modeled, and it was assumed that only two trains were operable (LAR Encl. 4-3, Ref. [23], Figure 5.8.1 Pg. 74). The multi-directional velocity vectors throughout containment during steady-state recirculation are indicative of homogenous mixing.

NOC-AE-14003103 Attachment 4 Page 8 of 19 Figure 5.8.1 - Vectors showing pool flow direction (Case 1)

In the second break simulation (Case 2) a break on the Loop C SI Pump Discharge Line was modeled assuming that only two trains were operable (LAR Encl. 4-3, Ref. [23],

Figure 5.8.17 Pg. 91). The location of the (Case 2) break is shown below in Figure 5.8.17. The multi-directional velocity vectors throughout containment during steady-state recirculation are indicative of homogenous mixing.

NOC-AE-14003103 Attachment 4 Page 9 of 19 Figure 5.8.17 - Vectors showing pool flow direction (Case 2)

Results of the Case 1 analysis (LAR Encl. 4-3, Ref. [23], Figure 5.8.20 Pg. 81) for small pieces of fiberglass are shown in Figure 5.8.7. Contours in red show regions with tumbling velocities large enough (?0.12 ft/s) to tumble sunken small fiberglass debris along the floor, while contours in blue are regions with velocities below this tumbling velocity (LAR Encl. 4-3, Ref. [65]). Contours in yellow show turbulent kinetic energy (TKE) large enough (?0.034 ft 2/s 2 ) to keep small fiberglass debris suspended (LAR Encl.

4-3, Ref. [65]).

NOC-AE-14003103 Attachment 4 Page 10 of 19 Figure 5.8.7 - TKE and velocity with limits set at suspension/tumbling of small pieces of fiberglass (Case 1)

Figure 5.8.7 shows that for a near-sump break (Case 1), there is not a flow path to the strainer that has TKE large enough to suspend settling of small pieces of fiber glass.

Therefore the assumption that small debris is homogenously mixed in the pool without settling when calculating the solutions of Equation 84 (LAR Encl. 4-3, Pg. 209) is valid.

Results of the Case 2 analysis (LAR Encl. 4-3, Ref. [23], Figure 5.8.20, Pg. 94) for small pieces of fiberglass are shown in figure 5.8.20. Contours in red, show regions with tumbling velocities large enough (>0.12 ft/s) to tumble sunken small fiberglass debris along the floor, while contours in blue are regions with velocities below this tumbling velocity (LAR Encl. 4-3, Ref. [65]). Contours in yellow show turbulent kinetic energy (TKE) large enough (>0.034 ft2/s 2 ) to keep small fiberglass debris suspended (LAR Encl.

4-3, Ref. [65]).

NOC-AE-14003103 Attachment 4 Page 11 of 19 Figure 5.8.20 - TKE and velocity with limits set at suspension/tumbling of small pieces of fiberglass (Case 2ý Figure 5.8.20 shows that for a far-sump break (Case 2) there is no flow path to the strainer that has TKE large enough to suspend settling small pieces of fiber glass. This is consistent with the results of Figure 5.8.7, and confirms the assumption that small debris can be treated as homogenously mixed in the pool without settling when calculating the solutions of Equation 84 (LAR Encl. 4-3, Pg. 209).

Results of the Case 1 analysis (LAR Encl. 4-3, Ref. [23], Figure 5.8.11 Pg. 85) for fiber fines are shown below in figure 5.8.11. Contours in yellow show turbulent kinetic energy (TKE) large enough (>8.2E-6 ft2/s 2 ) to keep fine fiberglass debris suspended (LAR Encl. 4-3, Ref. [65]).

NOC-AE-14003103 Attachment 4 Page 12 of 19 Figure 5.8.11 - TKE and velocity with limits set at suspension/tumbling of individual fibers (Case 1)

After recirculation (Figure 5.8.11) fiber and coatings fines are likely to mix more evenly throughout the pool due to the wide spread regions of TKE large enough to suspend and mix these fines (LAR Encl. 4-3, Ref. [23], Figure 5.8.11, Pg. 85).

Results of the Case 2 analysis (LAR Encl. 4-3, Ref. [23], Figure 5.8.25 Pg. 99) for fiber fines are shown in figure 5.8.25. Contours in yellow show turbulent kinetic energy (TKE) great enough (>8.2E-6 ft2/s 2) to keep fine fiberglass debris in suspension (LAR Encl. 4-3, Ref. [65]).

NOC-AE-14003103 Attachment 4 Page 13 of 19 Figure 5.8.25 - TKE and velocity with limits set at suspension/tumbling of individual fibers (Case 2)

Figure 5.8.25 shows similar results to those of Figure 5.8.11 and confirming that the TKE after recirculation is large enough to keep fiber fines homogenously mixed throughout the containment pool.

Results of the Case 1 analysis (LAR Encl. 4-3, Ref. [23], Figure 5.8.12 Pg. 86) for fine coatings chips are shown in figure 5.8.12. Contours in yellow show turbulent kinetic energy (TKE) large enough (?0.006 ft 2/s 2) to keep fine coatings debris suspended (LAR Encl. 4-3, Ref. [65]).

NOC-AE-14003103 Attachment 4 Page 14 of 19 Figure 5.8.12 - TKE and velocity with limits set at suspension/tumbling of fine paint chips (Case 1)

After recirculation fine chips would be likely to mix evenly throughout the pool because of the large regions of TKE are large enough to suspend and mix these fines (LAR Encl. 4-3, Ref. [23], Figure 5.8.12, Pg. 86).

Results of the Case 2 analysis (LAR Encl. 4-3, Ref. [23], Figure 5.8.26 Pg. 100) for fine coatings chips are shown in figure 5.8.26. Contours in yellow show turbulent kinetic energy (TKE) large enough (>0.006 ft 2 /S 2 ) to keep fine coatings debris suspended (LAR Encl. 4-3, Ref. [65]).

NOC-AE-14003103 Attachment 4 Page 15 of 19 Figure 5.8.26 - TKE and velocity with limits set at suspension/tumbling of fine paint chips (Case 2)

The data shown in Figure 5.8.26 confirms the discussion for Figure 5.8.12 that coating fines will mix homogenously throughout the pool because of their low TKE suspension limit.

The CFD results for the small and fine debris show that homogenous mixing is applicable for these debris sizes. Large debris such as intact blankets of insulation and large pieces were found not to transport under STP conditions (LAR Encl. 4-3, Ref. [23]).

NOC-AE-1 4003103 Attachment 4 Page 16 of 19 SSIB, Transport: RAI 11d (d) It appears that the x(t) function calculates that all debris is at the strainer at time=O and decreases as time progresses. This appears to be the reverse of the actual condition expected. Please explain.

STP Response:

The x(t) function is not in LAR Encl. 4-3 and is not used in CASA Grande. The original formula for x(t) describes the mass of debris in the pool rather than the mass of debris on the strainer, which explains the non-intuitive interpretation.

NOC-AE-1 4003103 Attachment 4 Page 17 of 19 SSIB, Transport: RAI 11e (e) Please explain the basis for the depletion rate.

STP Response:

The x(t) function and its associated debris depletion rate is not in LAR Encl. 4-3 and is not used in CASA Grande. The depletion rate of suspended pool debris implemented in CASA Grande is based on fiber filtration and shedding rates obtained from strainer module testing. 100% of fiber reaching the core is permanently retained.

NOC-AE-1 4003103 Attachment 4 Page 18 of 19 SSIB, Transport: RAI 13 Table 2.5.32 of Volume 6.2 includes values for small pieces of Microtherm. This is inconsistent with other statements in the submittal that Microtherm is assumed to fail as 100 percent fines. Please clarify if all Microtherm pieces fail as fines. Please state how small pieces of Microtherm are treated in STP's evaluation.

STP Response:

Microtherm is assumed to fail as 100% fines and is treated as such in the STP evaluation.

The discrepancy in Volume 6.2 has been entered in the STP corrective action program to track correction for future submittals.

NOC-AE-1 4003103 Attachment 4 Page 19 of 19 SSIB, NPSH and Degasification: RAI 29 Volume 1, Section 1.1, "Structured Information Process Flow," and Volume 3, Sections 2.2.28, "Pump Gas Limits," and 5.7.4, "Acceptance Criterion: Pump Gas Void Limits,"

describe the methodology for calculating NPSH margin. The submittal states that if the void fraction exceeds 2 percent that the scenario is recorded as a failure. It is not clear that NPSH Required (NPSHR) is corrected for degasification that may occur as fluid passes through the debris bed as recommended by RG 1.82, Revision 4, 'Water Sources for Long Term Recirculation Cooling Following a Loss-of-Coolant Accident," Revision 4, March 2012 (ADAMS Accession No. ML111330278). Please clarify whether NPSHR is corrected for the void fraction at the pump inlet. If the NPSHR is not corrected for the void fraction, please provide a justification.

STP Response:

Within CASA Grande, NPSHr is corrected for the void fraction at the pump inlet in accordance with RG 1.82, Revision 4. Additionally, as described in Assumption 8.i, (LAR Encl. 4-3) the void fraction at the pumps is conservatively assumed to be the same as the void fraction at the strainer, i.e. no credit is taken for bubble collapse.

NOC-AE-1 4003103 Attachment 5 Attachment 5 Response to STSB Request for Additional Information: RAI 1, 2, 3

NOC-AE-1 4003103 Attachment 5 Page 1 of 5 Technical Specifications Branch (STSB): RAI 1 The proposed LAR provides assumptions of partial flow reduction of certain ECCS equipment during DBA events. Please explain how the proposed LAR ECCS flow assumptions are affected when the Configuration Risk Management Program (CRMP)

Technical Specification (TS) Completion Times are applied to TS inoperable ECCS SSCs. Include in your discussion, how the related ECCS equipment PRA functionality (as defined in NEI 06-09, Revision 0, "Risk-Informed Technical Specifications Initiative 4b, Risk-Managed Technical Specifications (RMTS) Guidelines," November 2006 (ADAMS Accession No. ML12286A322)) is included in the analysis and how the analysis assumptions are programmatically included the CRMP.

STP Response:

The RAIs are directed toward the application of the STP CRMP and STP's Risk Managed Technical Specifications (RMTS). RMTS is currently unique to STP. As a pilot for a risk-informed approach to GSI-191 debris issues, STP did not rely on RMTS and does not consider RMTS implementation necessary for adoption of the risk-informed methodology for responding to debris issues as they may affect Technical Specification (TS) requirements.

STP assesses inoperable or nonconforming conditions in accordance with procedures based on the guidance of Part 9900 of the NRC Inspection Manual. This includes conditions that affect the systems within the scope of this licensing application that are supported by the sumps; i.e., ECCS and CSS. STP's expectation is that applying RMTS to calculate a risk-informed completion time (RICT) for an emergent debris related condition will be rare. Most emergent debris related conditions are resolved within the original "frontstop" completion time by prompt removal of the debris.

If a degraded or nonconforming condition related to debris affecting the function of the emergency sumps is discovered, it will be evaluated in accordance with Part 9900 as noted above. The Shift Manager of the affected unit will likely request a prompt operability evaluation by Engineering. Engineering would make a determination of whether the debris could significantly affect the evaluations described in this application by consideration of the nature, quantity and location of the debris. If Engineering determines the debris is adequately represented by the existing analysis, they will recommend that the sumps are operable, no TS action would apply and the condition would be resolved in accordance with the station's corrective action program. If Engineering recommends that the sump(s) should be considered inoperable, and the condition can be quantified in the CRMP, then a RICT may be calculated. The requirements for application of RMTS are incorporated into the individual Technical Specifications (TS) and in an Administrative Control Program TS 6.8.3.k, which invokes NEI 06-09, all of which are implemented by station procedures.

Typically, quantification of the risk with the degraded or nonconforming condition will be performed in accordance with station procedures by conservatively failing or otherwise adjusting affected functions in RICTCal, which is a database of thousands of pre-

NOC-AE-1 4003103 Attachment 5 Page 2 of 5 quantified PRA plant configurations. The quantification will represent the actual plant configuration with the inoperable components, including any other PRA modeled components or functions that are unavailable at that time. Thus, for the example of ECCS flow, the quantification will include the functions affected by debris effects, plus the unavailability of any other train of ECCS that is unavailable at the time. If the "PRA functional" allowance is applied, the quantification can force some components to be unavailable for specific initiating events, such as large break LOCAs, while retaining availability for other PRA initiators.

With regard to changes to PRA applications, the STP RICTCal database would be updated to take into account the impact of the debris effects related to the concerns raised GSI-191 applicable to equipment functions in a manner consistent with the current practices defined in the STP CRMP and programmatic update of the STP PRA.

No changes would be necessary for the process for providing the associated RICT to the plant staff.

NOC-AE-1 4003103 Attachment 5 Page 3 of 5 Technical Specifications Branch (STSB): RAI 2 The STP CRMP could allow continued power operation with a loss of a TS safety function for up to 30 days. Please explain how a TS loss of function, but PRA functional (as defined by NEI 06-09, Revision 0) ECCS SSC is addressed in the analysis.

STP Response:

The STP TS Bases describe the requirements governing PRA functionality and are excerpted below. The last bullet of the example list is relevant to debris related conditions. The response to RAI 1, above, also addresses the application of "PRA functional". The STP TS Bases guidance is consistent with Section 2.3 of NEI 06-09, Revision 0. The key factors are:

1. The CRMP allows application of RMTS for emergent situations where all trains of a function are inoperable, provided there is PRA functionality, but does not allow application of RMTS for a complete loss of function. If function is completely lost, the non-RMTS allowed outage time must be applied.

2. For conditions where PRA functionality may be applied, the risk from the configuration must be quantifiable using the PRA. For such cases, PRA inputs would be modified to obtain new configuration risk values consistent with STP existing procedures and processes.

STP TS Bases Excerpt:

Application of the CRMP will provide action for conditions where more than one train or channel of a function is inoperable. In accordance with NEI 06-09, a RICT (Risk-informed Completion Time) may not be applied for configurations where there is a complete loss of function or for pre-planned activities when all trains of equipment required by the TS LCO would be inoperable. It is permissible to apply a RICT for emergent conditions where all trains of equipment required by the LCO are inoperable provided one or more of the trains are functional as described in the guidance.

If a component is determined to be inoperable, it may still be considered to have PRA Functionality for calculation of a RICT if there is reasonable assurance that it can perform its required functions for events not affected by the degraded or non-conforming condition and if the condition can be quantified in the PRA. If these conditions are not met, the component will be assumed to be non-functional for calculating the RICT; i.e., it will have no PRA Functionality.

Examples of where a component has PRA Functionality such that the condition could be quantified in the determination of an allowed outage time are listed below:

• SSCs (Systems, Structures, Components) that don't meet seismic requirements but are otherwise capable of performing their design function.

NOC-AE-1 4003103 Attachment 5 Page 4 of 5

" SSCs that are inoperable but secured in their safe position (e.g., a closed containment isolation valve).

" SSCs powered from a source other than their normal power source, provided the alternate power source is modeled in the PRA.

" An SSC with an inoperable automatic function if the manual actuation of the SSC is modeled in the PRA (e.g., a diesel generator with an inoperable sequencer). Actuation channels are associated with their actuated components or trains. Loss of actuation channels is not considered a Loss of Function unless no train of the actuated SSC function has PRA Functionality.

" An SSC that is functional for mitigation of a set of events (e.g. steam generator tube rupture, small break LOCA) but is not functional for other events for which it is credited (e.g. large break LOCA or steam line break),

providing the PRA model can quantify the risk for the calculation of a RICT.

An example of this type of condition is degradation of environmental qualification.

NOC-AE-1 4003103 Attachment 5 Page 5 of 5 Technical Specifications Branch (STSB): RAI 3 Please explain how the assumptions and analysis for fibrous material impact on ECCS flow are verified and maintained programmatically (i.e., how and at what frequency are any physical or material changes to the analyzed impact zones evaluated and what physical or material changes would initiate a reevaluation of the affected impact zone).

STP Response:

Insulation replacement inside containment is either a like-for-like replacement as a maintenance activity ("rework") or is a modification with a design change that has been approved by STPNOC Engineering.

The STPNOC design change process procedures ensure that new insulation material that differs from the initial design is evaluated. The STPNOC design change process also calls for evaluations of added metals such as aluminum that could contribute to post-LOCA chemical effects in the sump water. The process looks at coatings that are to be used inside containment. Impacts to post-LOCA recirculation flow paths and recirculation sump debris impact on internals of fluid containing components are part of the design change evaluation process described in the procedure.

Examples of where reevaluation of the impact zone would be warranted include:

• Introduction of a new type of material

• Increase of existing fibrous material

• Increase of existing aluminum material "Procedures and Activities in the Licensing Basis" in Part I of Enclosure 4-1 and Section 3.3.4 of Enclosure 3 to Reference 1 of the cover letter provide additional information.