ML20217C257

Review of Neutron Fluence Data for Palisades Reactor Pressure Vessel

ML20217C257
Person / Time
Site:	Palisades
Issue date:	12/16/1997
From:	Hopkins W BECHTEL POWER CORP.
To:
Shared Package
ML18068A330	List: ML20217C257 ML20217C274 ML20217C310
References
NUDOCS 9804230255
Download: ML20217C257 (28)

Text

12/16/97 REVIEW OF THE NFUTRON FLUENCE DATA FOR THE PALISADES REACTOR PRESSURE VESSEL W. C. Ilopkins Bechtel Power Corporation l

Gaithersburg, Maryland l

1.

DESCRIPTION The reactor pressure vessel for the 2530 MWt Palisades Nuclear Power Plant was fabricated by Combustion Engineering. The plate material is A302B[1]. The axial vessel welds situated at 15 degrees from datum are the limiting vessel material for pressurized thermal shock [2] and were fabricated under Heat No. W5214 (%Cu = 0.21, %Ni = 1.020[1]).

2.

OPERATING HISTORY [3]

The reactor has operated within normal system temperature and pressure ratings since initial power. The fast neutron Duence that governs the reactor embrittlement rate was changed in Cycle 8 when Palisades changed the fuel cycle to achieve the lowering of the fast neutron fluence on the reactor pressure vessel. The Duence rates for cycles 8 through 11 are based on funher reduction of the fast neutron fluence through a series of changes from a low leakage loading pattern to a very low pattern established through the use of thrice burned assemblies with steel rods substituted for normal fuel rods in selected assemblies on the outer core periphery. This low leakage fuel cycle strategy resulted in a projected end of life in the year l

2011 A.D. based on a fast neutron Buence of 1.32 x 10" neutrons /cm' from Cycle 11 continuing through end of life at a 85% plant capacity factor [3]. Currently the licensed life of the plant is 2007 A.D.

3.

DISCUSSION OF PRIMARY ISSUE The primary issue is to assure that the reactor pressure vessel integrity is maintained for all design basis conditions. The primary design basis condition in question is " Pressurized Thermal Shock (FTS)" [2]. Section 50.61 of 10CFR50 [2] requires that the reactor pressure vessel must 1

9804230255 980420 PDR ADOCK 05000255 P

PDR

12/16/97 maintain a toughness throughout its life so that if an accidental cooling transient should occur, the vessel's integrity will be uncompromised. During its life the vessel wall is embrittled by the high energy neutrons from the core. Depending on trace elements such as copper and nickel in the steel, the irradiation of the vessel's steel microstructure can result in unacceptable embrittlement. The degree of embrittlement is characterized by the ductile-to-brittle transition temperature, RT,, i.e. the temperature below which the steel behaves in a brittle manner. The Charpy "V" notch test for steel can be used to define transition temperature.

(

Section 50.61 of 10CFR.50 [2] sets fracture toughness requirements so that reactor beltline materials are protected against PTS events. It does this through a set of empirical correlations based on fast neutron fluence and the copper and nickel content of the steel plate or weld used.

If the transition temperature adjusted for irradiation exceeds 270 F for vertical welds or 300 F l

for circumferential welds, the licensee is to perform an analysis showing how the reactor can be i

safely operated with these high transition temperatures. USNRC Regulatory Guide 1.99 [4]

l gives funher guidance on how to evaluate whether a pressure vessel integrity is acceptable based on an empirical formula. One of the key variables for that formula is the total fast (Energy > 1 Mev) neutron exposure or fast neutron fluence that the vessel is expected to receive

[

over its lifetime.

The Consumers low leakage strategy outlined in section 2 has resulted in an overall projected fast neutron fluence reduction. Furthermore, Palisades has done a systematic evaluation [3] of the copper and nickel variability of the weld material in response to the USNRC Generic Letter 92-02 that was a more rigorous evaluation than that originally reported in the original assessment of the life of the vessel. As a result, the currently docketed analysis [5] predicts that the Palisades reactor pressure vessel could still operate safely until 2011 A. D.

Consumers Energy has pursued a vigorous program to establish a best estimate value for the fast neutron fluence which involves both state of the art calculational methods as well as experimental methods. This program has been closely scmtinized by the US Nuclear Regulatory Commission and has resulted in ever increasing ref'mements to the best value of the fast neutron fluence. In April,1996 Consumers submitted their updated reactor vessel fluences described _in Section 2 above to the USNRC for review and approval [3]. On December 20, 1996 the NRC issued an interim Safety Evaluation Report [6] that allowed only the 8%

j 2

)

12/16/97 reduction in the value of the fast neutron fluence attributable to the changes in physical plant parameters as opposed to the 25?c requested in the April Consumer's submittal. The remaining 17% is ascribed to:

127c due to bias in the reactor cavity and vessel dosimeter measurements 57c due to spectral adjustments of the dosimeter measurements Over a period of several months Consumers met with the NRC staff and replied on the docket to various Request for Additional Information (RAIs) to address these two issues [5]. At this i

date no final resolution has been reached. The two bulleted issues above are the current items of contention and will be discussed in the next section i

l 4.

OBSERVATIONS & DISCUSSION The docketed material presented by Consumers meets the guidance set forth in the Draft USNRC Regulatory Guide DG-1053, " Calculational and Dosimetry Methods for Determining Pressure Vessel Neutron Fluence" as clarified by the comment; of the Nuclear Energy J

Institute's Fluence & Dosimetry Task Force [7] as well as the guidance set forth in the set of ASTM Standards and Guides in the E706 Standard Master Matrix of Light Water Reactor Pressure Vessel Surveillance Standards [8].

With respect to the issues raised by the NRC shown in Section 3, the following observations can be made:

The issue of the 12% bias in the cavity vs. the in-vessel measurements has been well addressed by Consumers in their docketed reply [5]. That the ratio of the measured to calculated values (M/C) of the fast neutron fluence ranges from 0.830 for the cavity to 0.835 for the in-vessel ratio shows good statistical agreement for the data base of 34 samples.

The issue of whether there is a "spectnim dependent error in the DORT calculations"[6] has been addressed also in [5] to the effect that similar observations particularly in the reaction rates for Fe-54 and Ni-58 dosimetry have been obsened by others (9,10] as well as within the generic Westinghouse dosimetry data base.

'I 3

I 12/16/97 One of the most dramatic demonstrations of how well nuclear technology has done in analyzing the Palisades reacior vessel is to observe how closely the calculated results compare among the various independent calculators. The Consumers / Westinghouse DORT[1il results compare very well not only to the DORT calculations independently done by the NRC's contractor at Brookhaven National Laboratory but also to the Monte Carlo calculations performed by AEA Technology Engineering Services with the three dimensional, point energy group code MCBEND.

TABLE 1 l

Comparison of Code Calculated Values for Fast Neutron Fluence at End of Cycle 11 Calculator Valve CONSUMERS / WESTINGHOUSE 1.59E+19 USNRC/BROOKHAVEN NATIONAL LAB 1.60E+19 t

AEA TECHNOLOGY ENGINEERING 1.52E+19

)

SERVICES 5.

RECOMMENDATIONS Table I suggests that for such a tight fit of the calculated fast neutron fluence values, the apparent. correction necessitated by the measurements as recommended by Consumers / Westinghouse (M/C ratio = 0.88[cf. 4]) may be due to a bias in the basic data which is input to the transport code calculated numbers and not the codes themselves. A discussien of some of these basic parameters were suggested in the last Consumers docket [5] and a repeat of all those will not be discussed here. Only the ones that are viewed as needing an independent reassessment or re-evaluation are discussed.

1. Density of the A302B Steel A review of the data supplied in the Consumers docketed submittal of April, 1996[3] contained a table of material densities and compositions in the Appendix containing the AEA analysis with MCBEND. That table showed a density of 7.9 4

12/16/97 grams /cc was used. References [13] indicate that mild carbon steels can range down to densities of 7.64 gm/cc. This suggests a quick review of the material records for the Palisades, specifically, the Combustion Engineering Shop Traveler Records that

)

were part of the Consumers submittal [5] for August 27, 1996 which contain the dimensions of the vessel and should contain the net weight for that vessel section.

2. Power Gradient in the Outer Assemblies Of all the parameters discussed as possible sources of the bias in the dialogue between Consumers and the NRC, this item stands out as the one needing an independent assessment outside of both Consumers and its fuel supplier. Estimates of the uncenainty for the thrice burned fuel assemblies with the stainless steel dummy rods have ranged as high as 5 to 10%.

3. Independent Re-evaluation of the Dosimetry Although this would be a big undertaking, an independent re-evaluation of the existing dosimetry data with another computer code that is similar to the FERRET code [14]

used in the standard Westinghouse methodology by an independent third party would go far in establishing a more rigorous credibility with the NRC reviewers. One such possible code exists in the EPRI sponsored suite of codes called LEPRICON[15]. This system, designed to be the ultimate in assessing and adjusting both measurements and calculations within a least squares statistical framework, was to estab'ish insight into those parameters which were driving the uncertainty in the evaluation of the "best estimate" for the fast neutron fluence and could serve as a reference for other methods which did not need to be so comprehensive from a research and development perspective. All of the modules of LEPRICON do not have to be run to do an independent assessment of the spectral unfolding capability of FERRET, rather the one module that was developed for just that part of the overall adjustment process could be used.

5

12/16/97 6.

REFERENCES

1. USNRC Reactor Vessel Integrity Data Base -- Summary File for PTS, July 22,1995.

2. Title 10, Part 50.61, Code of Federal Regulatin_ns.

n

3. Consumers Power Submittal to USNRC, April 4,1996.

4. Regulatory Guide 1.99, Radiation Embrittlement of Reactor Materials, Revision 2.

USNRC, May 1988.

5. Consumers Power Submittal to USNRC, June 26,1997.

I

6. Letter from J.H. Hannon (NRC) to T. C. Bordine ((Consumers), December 20,1996.

7. Letter from T. E. Tipton (NEI) to M. E. Mayfield (USNRC), August 30,1996.

)

8."E706 Standard Master Matrix of Light Water Reactor Pressure Vessel Surveillance Standards," ASTM Annual Book of Standards, Volume 12.02, Philadelphia,1994.

)

9. I. Remec, and F. B. K. Kam, "H. B. Robinson-2 Pressure Vessel Benchmark,"

NUREG/CR-6453 (DRAFTL to be published.

10. A. Bevilaqua, et al., "Special Dosimetry at Saint Laurent B1 MOX-Loaded Unit,"

Proceedings of the Eighth ASTM-Euratom Symposium on Reactor Dosimetry. pp.132, Vail, Colorado, September 1993.

I1.DORT. Two Dimensional Discrete Ordinates Transport. RSIC Computer Code Collection, CCC-484 Oak Ridge National Laboratory,1988.

12. MCBEND -- User's Guide for Version 9A, ANSWERS /MCBEND(94)l5.

13. Handbook of Chemistry and Physics. 44th Edition, The Chemical Rubber Publishing i

Company,1961, pp.1532-1533.

14. F. Smittroth, " FERRET: Least-squares Solution to Nuclear DATA and Reactor Physics Problems," HEDL-TME-79-40, Hanford Engineering Laboratories,1979,

15. R. E. Maerker, J. J. Wagschal, and B. L. Broadhead, " Development and Demonstration of an Advanced Methodology for LWR Dosimetry. EPRI NP-2188.1981.

6

1

)

l ENCLOSURE 3 CONSUMERS ENERGY COMPANY PALISADES PLANT DOCKET 50-255 I

)

Palisades Reactor Vessel Fluence J

Resume & Independent Review Completed by Richard Maerker Oak Ridge National Laboratories, Retired i

RESUMb 0F RICHARD MAERKER Born: Jackson, Michigan on July 9,1928 B.S. University of Tennessee 1949 in Engineering Physics M.S.

University of Tennessee 1951 in Physics Ph.D. University of Tennessee 1953 in Nuclear Physics Teaching Assistant in Physics, University of Tennessee, 1952 Member, Research Staff, Eastman Kodak Research Laboratory, Rochester, N.Y.,

1953 Member, Research Staff, University of Tennessee Physics Dept., 1954,55 Assistant Professor Physics, Tulane University, 1956,57,58 Research Associate, Tulane University, 1956,57,58,59. Exterior Ballis-tics of Rockets for Redstone Arsenal Head, Shielding Group, Oak Ridge National Laboratory, 1959-62 Member, Reactor Shielding Group, Fontenay-Aux-Roses, Centre d' Etudes Nucleaire, Paris, France 1962,63 Shielding Group, Neutron Physics Division, Oak Ridge National Labora-tory, 1963-93 Retired: 1993-Present Author of over 100 publications, over ten of which appeared in Nuclear Science and Engineering. Member of CSWEG 1978-85. Presented papers at ANS annual and winter meetings throughout career and contributed papers and chaired sessions at ASTM-Euratom Symposia in Washington, Ispra, Italy, Jackson Hole, Hamburg, Germany, Vail, and Strasbourg, France. Visited Winfrith and other laboratories in England on several

)

occasions, as well as Nuclear Laboratories in Sweden and Belgium.

Chief Investigator on work performed at ORNL and funded by EPRI and the NRC on development of the LEPRICON code system, 1980-90.

i

f l

i l

March 4, 1998 John Kneeland Palis.. des Nuclear Plant 27780 Blue Star Memorial Highway Covert, MI 49043 This is a revised summary of my impressions of the meeting held at ORNL on Monday, January 5, 1998, involving Palisades, Westing-house, ORNL personnel and myself. This summary and the meeting were requested by Ross Snuggerud to provide a forum for answering questions which I had raised in my letter report of November 25, 1997 as well as to enable a general discussion to be held among the various parties responsible for providing the Palisades data to the NRC to assess and comment upon the method used by Westing-house to satisfy the requirements of the PTS rule.

Ross covered many of the questions that I had raised and I was satisfied with his answers. I also felt I had a better understand.

ing of the FERRET code after listening to Stan Anderson's explan-ation of how he used it. However, and this is indeed unfortunate, I failed to instigate a discussion of the more than just a few instances of illogical, inconsistent, and what I feel to be in-correct behavior of the FERRET results which I had pointed out in the continuing paragraph on pages 12 and 13 of my report. I still strongly feel, therefore, that not only are the FERRET cavity ad-justments inadequate in providing any useful information concern-ing the accuracy of the calculated fluxes at the PV inner radius for reasons presented in the report, but because of their some-times spurious behavior may be of questionable accuracy in the cavity locations as well until these questions are resolved.

This is not to say, however, that I feel that the method of least squares is not the preferred approach to the problem. But in or-de,r for it to provide credible anc reliable results it must be flexible enough to differentiate between in-vessel and ex-vessel dosimetry locations and have the ability to translate the effects of parameter adjustments to other important flux locations that are inaccessible to dosimetry. To ensure maximum credibility, the procedure should be capable of incorporating into the LWR analy-l sis the M/C comparisons of some simpler benchmark experiments

(

sharing sensitivities to important parameters that are common to the LWR, and then of performing a simultaneous global adj ustment of all the data. In this manner, the generic data such as iron cross sections are adjusted in a fashion consistent with all the transport calculations involved, leaving the LWR plant specific 1

data such as core source, dosimetry locations and PV thickness to be adj usted more independently, with less interference from eff-ects arising from inaccuracies in the generic data. To maintain consistency thro u g hout i t goes without saying that the use of the same cross sections and transport methods in the calculations of both the benchmarks and LWR is sine qua non to the success of the least-squares procedure. All the input data (correlations, stan-dard deviations, and values of M and C for example) as well as precalculated group flux sensitivities to the significant parame-ters in the transport calculation should be as accurate as possi-ble, and the calculated and measured dosimetry should fall within about 30% of each other. Thus the calculations should be s t a t e-o f-the-art to ensure that the parameter adj ustments will be small and the sensitivities will remain linear over the range of the adj ustments.

If the above least-square procedures are adopted and the caveats heeded, on the average the parameter adjustments will be closely limited to their standard deviations. This, the Chi-square per degree of freedom test, can be applied to the data set even prior to adj ustment to test the consistency of the set, and provides a strong measure of security that the set is self-consistent if the value of Chi-square is close to unity. Generally speaking, the equal importance of the measurements and the calculations in the procedure must be acknowledged, as is the fact that, by vir-tue of the adj ustments to a more consistent set, all uncertain-ties are decreased from their original values.

For all these reasons, since there are uncertainties associated with all data employed in reactor analysis, it is incumbent upon the analyst to rely on such a least-squares procedure to further improve his knowledge of the accuracy of his calculations and to draw defensible conclusions about their significance and the ram-ifications thereof.

Previous to this meeting I had asked Igor Remec, who had authored the two reports on the PCA and HB Robinson-2 Cycle 9 benchmarks, to provide any results he might have on the magnitude of the error incurred in neglecting effects of a finite axial core. He managed to come up with some calculated data using the SAILOR li-brary based on ENDF/B-IV for a two-loop 1800MWt reactor having a 6.761n.-thick pressure vessel. For the case of near-midplane loc-ations, and using the flux synthesis procedure alluded to in my report which involves RZ and R transport calculations in addition I

to R8, the effect of neglecting the finiteness of the axial core j

source at the PV inner wall and in the cavity was found to be overpredictions of 2.6% and 4.5% respectively, based on compari-sons of six dosimeter responses. Since the Palisades PV is some-what thicker, we estimate the effect in the Palisades cavity to be around 5%.

i 2

1 Notice that the 2.6% compares very favorably with the 3% that West-inghouse overcalculated the AEA Monte Carlo results at the PV inner radius. Staying clear of any FERRET adj ustment results and relying on the M/C values only, the estimate of 5% in the Palisade.= cavity dos-imetry overprediction would leave 13%-5%=8% as still unaccounted for.

For the in-vessel dosimetry the overprediction still unaccounted for is 8%-3%=5% as before, and one can argue persuasively that this is very likely due to inaccuracy in the calculated core leakage source.

If it is (Ross remarked that the source uncertainty cannot be reduced by normalizing to measured in-core assembly powers because these measurements are unavailable), then this renormalization of the cal-culation would affect the cavity dosimetry as well. There is also ev-idence that the venturi problem alone would contribute half of this overprediction by itself, but Ross replied it has not yet been taken into account in the calculations. If it were, this would leave merely a 2.5% overprediction due to source leakage ina: curacies, which is even further within the estimated uncertainty of 7% that it is known.

The result of these considerations, now revised from the one postula-ted in my report, would be that a 5% overprediction in the source, or alternatively shared between the source and power discrepancy, is probable, but only that part due to the source leakage can be j usti-fied as a valid reason for extending the operating lifetime until the PTS screening criterion is met.

Without using any adj ustment procedure results, but incorporating the corrections to the calculated fluxes of -3% at the PV inner radius and -5% at all cavity locations followed by a renormalization by an-other -5% to both locations would result in M/C values of in-vessel dosimetry averaging 1.0 and of cavity dosimetry averaging 0.97.

One remaining apparent discrepancy should be reconciled before any definitive conclusions are reached, and that is the accuracy of the BUGLE-93 iron cross sections. Certainly, they are far superior to any of the earlier evaluations based on ENDF/B-IV. However, Remec's PCA and HB Robinson-2 cycle 9 benchmark calculations using the same cross section sets for both transport and dosimetry as were used for Pali-sades, with all three being corrected for source finite axial effects, yield M/C dosimetry values which average 1.04 and 1.09 at the PV in-ner radius and 3T/4 locations respectively for the PCA, and 1.10 for both PV inner radius and cavity locations for HBR-2. These ratios are from 4% to 13% higher than the corresponding Palisades values, and 1ndicate that there may still be inaccuracies in the source levels in one or moreoof these calculations combined with relatively small re-maining inaccuracies in the iron cross sections. This issue was not pursued at the meeting, but only mentioned in passing. No definitive conclusions could be reached about the status of the request to NRC for lifetime extension based on this work, but the above analysis does differ in several important respects from the one offered in my letter report.

W 3

s LIST OF ATTENDEES Ross Snuggerud, Palisades Stan Anderson, Westinghouse John Perock, Westinghouse Joe Pace, ORNL Igor Remec, ORNL Richard Maerker, Consultant 1

i 4

l l

November 25,1997 Ross D. Snuggerud Palisades Nuclear Plant 27780 Blue Star Memorial Highway Covert, MI 49043 I have read all the materials that were sent to me and found them informative and,for the most part, clearly and professionally presented, There are essentially three separate parts to the method chosen by Palisades for satisfying the requirements consistent with the basis of 10CFR50.61, the PTS rule. They are (1) accurate transport calcula-tions of cycle-dependent neutron and gamma-ray fluences using the method of discrete ordinates with the BUGLE-93 cross-section library, (2) state-of-the-art dosimetry measurements performed at both in-vessel and ex-vessel locations and (3) combining of the measured and calculated dosimeter reaction rates to yield important information concerning the magnitude of the accumulated fluences, since on-line start-up, at points throughout the pressure vessel (PV). Of these three steps. I have found no reason to question the accuracy of the measurements. My remarks will therefore be directed almost exclusively to the remaining two.

The first part of this report will concern itself with comments and j

suggestions on the subject matter addressed in the letter dated April 4, 1996, together with its attachments, from Richard W.

Smedley to the USNRC.

The method adopted by Westinghouse for performing the transport calcu-4 t e x t b o o k '.' H o w e v e r, I feel compelled lations-is generally adequate-and to make several comments regarding them.

The core source was based entirely on a diffusion theory calculation l

using SIMULATE-3, which I assume to be a three-dimensional code. This is fine for obtaining the relative pin powers, but aren't there in-core instrumentation data available to renormalize these calculations i

to the various assembly powers? This ties the calculations to reality and provides a basis for estimating the source uncertainties as well.

It is my recollection from the work of M.L.

Williams, now at LSU, that based on analysis of several core benchmarks performed at the VENUS facility in Mol, Belgium, his conclusion was that disagreements of up to 7% could readily exist between the measurements, diffusion theory, and transport theory in the powers of the peripheral elements. (I not-iced that further along in the attachments that an uncertainty of 7 or 8% has been estimated for the source. My point is that this can and should be reduced, for it is not an insignificant effect.)

1

l l

l At the bottom of pg.1 I believe there is an important typo in which "ENDF/B-VI" was mistakenly substituted for "ENDF/B-IV". It is my understanding that the most significant difference between the two l

versions of the iron cross sections lies in the total inelastic scat-tering cross sections which are about 6% lower in the range 3 to 8Mev in the new evaluation relative to the old. The reduction in this cross j

section results in less average energy loss per scattering event with l

a corresponding increase in the transmission of these higher energy L

neutrons through the steels comprising the internal reactor structures l

and the PV. The change toward more forward scattering is less import-ant.

J l

I assume that the reason for quarter core symmetry adopted in the cal-culations was to gauge the effects of the dosimetry capsules and their holders in shielding the PV. There was no explanation given for not using octal symmetry. The results in Tables 2.2.1 through 2.2.3 do indicate a slight octal asymmetry dependence on cycle which.should rule out any effect caused by the placement of the in-vessel dosimetry.

J Perhaps the reason is a cycle dependent core loading asymmetry in the peripheral assemblies.

In Table 2.2.3, although the Monte Carlo statistics are not given, I think there is significance in the last column labeled "% Difference AEA to New" to the extent that all. azimuthal entries are negative, be-cause there may be a good reason for it. Because the core is finite as well in the axial direction, the effect cannot be calculated using 2-D geometry, since the 2-D calculation must perforce assume a vertically infinite core. The effect of a finite axial core is to produce fluxes which decrease with increasing distance from the core relative to those from an infinite core. At the PV inner radius the effect may well be responsible for the average value of about -3% indicated.

Again, if one averages all of the cycle 8 and 9 (M/C) values calcula-ted in the cavity by Westinghouse in Table 7.2-1 and compares them with the (M/C) values calculated by AEA (after the latter have been decreased by 18% to correct for an additional 0.31n. in the PV thick-ness, see discussion in Section 7.4, pg. 9 of attachment 3), one obtains 0.87 for the Westinghouse calculations and (1/0.96) = 1.04 for the Monte Carlo calculations. I submit that at least a good fraction of this 18 % dif f erence is due to finite axial core effect. It is poss-ible to use a flux synthesis procedure involving two additional but simpler discrete ordinates calculations that corrects the RO fluxes

• for this effect, and is very accurate if the azimuthal and axial flux distributions are independent of each other. This procedure was used i

in the calculations.of the HBR-2 benchmark, and resulted in (M/C) values in the cavity averaging 1.10 Commencing in Section 2.3, the phrase " Capsule Measurement" is a mis-nomer that can lead to great confusion. What are being described are on the mismatches be-adjusted calculated fluences based in some sense tween the calculated and measured dosimetry reaction rates. The meas-urements are not measures of fluences or fluxes; they are simply guide posts to confirm the accuracy of the calculations. The use of " capsule measurements" should be limited to the original (i.e.,

unadjusted) 2

i activities only. The heading for Table 2.3.1 should be changed to l

" Changes in Adjusted Capsule Fluxes Lying Above 1Mev", etc. It would be of interest to provide a. table somewhere in attachment 1 which l

shows the comparisons of the measurements with the new calculations and the old calculations. It does surprise me that the adjusted cal-culated cavity capsule fluxes in Table 2.3.1 are less for the new calculations compared to the old since the version VI iron cross sec-l tions will produce higher cavity fluxes than version IV and, histori-cally, are in better agreement with cavity dosimetry. This interpret-l ation would seem to indicate that the earlier calculations agreed j

better with the cavity measurements than the current ones. My experi-ence with HBR-2, cycle 9, using version IV cross sections yielded M/C dosimetry comparisons in the cavity in the neighborhood of 1.49, whereas changing only the library to BUGLE-93,Igor Remek has obtained values of M/C averaging 1.10. Similar conclusions were obtained from analysis of the PCA 12/13 benchmark, in which version IV cross sec-tions produced M/C values averaging 1,43 at the 3T/4 location and the BUGLE-93 library values of 1.10 again. In the earlier calculation the M/C values increased with increasing penetration into the PV mockup, indicating a possible problem with the iron cross sections above 3Mev, but the BUGLE-93 results showed no such fall off in the calculations.

Notice that slight underpredictions of the cavity or near-cavity dos-imetry still occur in these benchmarks even using BUGLE-93. Because of the' absence of a thermal shield in Palisades there are several inches less equivalent iron in the neutron paths from the core into i

the cavity than in HBR-2 and slightly more iron than in the PCA mockup. It therefore seems reasonable that the BUGLE-93 library should still tend to underpredict the Palisades cavity dosimetry measurements by.around 10%. An inspection of Table 7.2-1 in attach-ment 2 shows that the unadjusted calculations tend to overpredict the cavity dosimetry measurements by about 13% on the average, depending to some extent on the particular cycle, location, and dosimeter. This discrepancy in the cavity between overprediction in Palisades and underprediction in two reputable benchmarks should be resolved, for cavity dosimetry can yield important information about cross-section deficiencies and effects of transport method approximations which can affect the analysis of in-vessel dosimetry as well as PV fluences.

When the cavity dosimetry comparisons of the revised AEA Monte Carlo I

Pelisades calculations are considered as well, they are consistent with the trend of a modest underprediction. Since the calculations of HBR-2 and the PCA by ORNL and of Palisades by AEA all contained cor-rections for effects of a finite axial core and Palisades by Westing-i house did not, it seems logical to ascribe at least some of the cavity overprediction to the neglect of these finite core effects.

There thus is strong evidence that the two principal causes of the overprediction in the Westinghouse calculations are the core source and the assumption of an infinite axial core. Based on the (M/C) com-parisons of the in-vessel dosimetry which average about 0.92 (see Table 7.2-1) an estimate of the inaccuracy in the source term is that it is about 5% high, which is well within the estimated uncertainty that.it is known. This component is constant throughout the PV and cavity as well since it essentially represents a flux renormalization.

3

The 8% high in-vessel calculation thus arises from a 5% high source and a 3% high effect from assuming an infinite axial core. In the cavity, on the other. hand, the fluxes are still overcalculated by 8%

after subtracting off the 5% source contribution. If we ascribe the f ull 18 % disagreement between the AEA and Westinghouse calculations to the finite source we end up with the estimate that the corrected i

Westinghouse fluxes would now be about 23% lower in the cavity than l'

Westinghouse calculated them to be. Since these uncorrected fluxes were 13% too high based on the dosimetry comparisons, the combined effect of the source renormalization and its finite axial extension would lead to an underprediction of the cavity dosiraetry of around 10%.

This underprediction is in perfect agreement with the HBR-2 and PCA analyses and within 5% of the AEA Palisades revised analysis.

If the above analysis is correct, then there seems to be good evi-dence that.the iron cross sections are responsible for the underesti-mation of the cavity calculations in the neighborhood of 10%. If this is due to the need for a further 2% or so decrease in the inelastic scattering cross section or something else is problematical at the present time. Measurements of these cross sections have accuracies of somewhere in the vicinity of only 5%. The absence of a thermal shield in Palisades should not affect the above argument since it has been demonstrated that these calculations can predict fast-neutron attenu-ation through nine inches of carbon steel to within about 10%. Any error incurred in the transmission through a thermal shield would be virtually negligible in HBR-2 or the PCA. The only effect of an absent thermal shield on these calculations is the enhancement of the photofission contributions to the in-vessel fission dosimeters.

The following comments are made on material appearing in Attachment 2.

On page 3-14 line 1: There is no dosimetry file as such in ENDF/B-VI;instead, dosimetry reaction cross sections appear individually in the isotopic files.

Lines 8,9: We have found that on occasion correlations can be very important. As a matter of fact, the correlations in the group fluxes at the dosimeter locations and the PV group fluxes are at the heart of any defensible adjustment procedure. These correlations which arise as a result of sensitivity to changes in a given cross section,

'for example, allow one to obtain adj ustments in the PV fluxes as a direct result of adjustments in the fluxes at cavity locations.

This will be discussed in greater detail later on in this letter.

Pages 3-14 through 3-17: In our development of the LEPRICON adjus:-

ment code, many sources of uncertainty that can contribute to flux uncertainties are identified and quantified. These included iron partial cross sections, source normalization, plant structure toler-ances, coolant densities, biological concrete shield compositions and 1

water content, absolute locations of in-vessel dosimetry, and method approximations such as 3-dimensional flux synthesis, etc. This survey concluded that the major contributor was the iron total inelastic cross section because of high flux sensitivity to it,with still 4

j

important but somewhat less cor.cributions from source normalization, l

the absolute locations of the in-vessel dosimetry, and the tolerances in the PV inner radius and its tnickness. Uncertainties in the flux synthesis procedure were determined by comparison with Monte Carlo and found to lie within the Monte Carlo statistics of about 3% and hence were not an important contributor. This survey was conducted in order to establish the main causes of the poor agreement between calculations with ENDF/IV cross sections and dosimetry measurements in the PCA and later on in ANO-1 Unit 1 and HBR-2.

What determines the importance of a given source of uncertainty is the size of its uncertainty and the sensitivity of the group fluxes at important locations to changes in the source. These sensitivities can be accurately calculated by the use of adjoint formalisms in the case of nuclear data or perurbat' ions in 1-D forward transport runs for most of the others. Multiplying the sensitivities (in the form % change in the group flux per % change in the source of uncer-tainty) by the uncertainties in the various sources yields the group flux uncertainties, thus giving rise to correlations in the flux I

uncertainties. We then say the two fluxes are correlated through the particular source of uncertainty (i.e.,

parameter) chosen. Per-I forming these multiplications for all the parameters considered and then summing them up results in uncertainties for and correlations f

between all the group fluxes at one location. The procedure can be

{

readily extended to obtaining correlations between group fluxes j

l at two locations such as a dosimeter and the PV inner radius, which l

has been pointed out a little earlier. The procedure used in FERRET to obtain source specific group flux correlations and uncertainties l

is missing. It does not identify the various sources of uncertainty or consider the group flux sensitivities to these sources. Instead, it constructs a mathematical expression for the flux correlations, and the parameters in the expression are set as input and apply to I

all sources of uncertainty equally, since there is nothing to dis-tinguish the correlations due to different sources of uncertainty.

Consider, for example, the flux correlations that arise from two different sources of uncertainty: the core source and inelastic iron cross section. The flux correlations from uncertainties in the core source a e all

+1, i.e.,

all the group fluxes are perfectly cor-related. The flux correlations from inelastic scattering in iron are dependent on the locations of the flux points. For in-vessel loca-tions, iron sensitivities are small and the correlations among the fluxes also small. For cavity locations, iron sensitivities are large and the correlations among the fluxes also large for cross section uncertainties of the order of a few percent or more. The correlations between the group fluxes at a cavity location and those inside the PV vary from small at the inner radius to large at the outer radius.

Thus the effect of inaccurate iron cross sections determined from the cavity dosimetry has little effect on the flux at the PV inner radius and the main source of correlation in the fluxes at the two locations comes from the core source uncertainty. This is consistent with the discussion presented earlier in which the adjustments to the latest Palisades calculations perhaps involve these two sources of uncertainty along with a third, the finite axial core effect, whose flux correlations between cavity and inner radius locations are very similar to those induced by the iron cross sections.

5

I confess I do not understand the FERRET code very well, since there is little detailed explanatory literature on it, so some of my remarks about what it purports to do and how it does it must be tempered with this in mind. My feeling is that there has not been paid enough atten-tion to the details I have just introduced that are necessary for any defensible adjustment procedure that involves translating the results of a reconciliation procedure at a dosimetry location to flux adj ust-ments at another location. There is nothing available after the ad-justment at the dosimeter location to identify the contributions from the individual sources of uncertainty so that they may be used with their adjustments and their sensitivities to calculate changes in the fluxes along with revised uncertainties at other locations. It becomes a mathematical exercise with little touch with reality.

Moreover, following the discussion in Section 3.3, I fail to see how the FERRET calculation is used to determine the "Best Estimate" react-or vessel exposure. It is probably wrapped up in the factor K, but K is simply defined as "the plant specific (M/C) derived from all avail-able surveillance capsule and reactor cavity dosimetry data". Perhaps FERRET is used to weight the various (M/C) values, but it is not clear to me how this is accomplished. Although I am probably wrong, it seems that a single weighted average of all the (M/C) values is used to ad-just the calculated fluxes everywhere. If so, this amounts to a simple renormalization, which as we have seen is incorrect. If the reason for the mismatch in the cavity dosimetry is because of neglect of effects of a finite axial core and somewhat still inaccurate iron cross sec-tions, then the effect on the PV inner radius fluxes is very small.

The only component that mutually affects both the dosimeter and PV inner radius locations is the core source uncertainty, and it is a relatively minor component in the flux uncertainties at the dosimeter although a major one at the PV inner radius. An adjustment procedure to be reliable must treat each of the sources of uncertainty in suf-ficient detail to properly evaluate its contribution to group flux ad-justments at other locations.

In Tables 4.1-1 through 4.1-4 the factors F are undefined and the gamma-ray spectra are not shown. Also in Table 4.1-4,for which cycles do these spectra represent? In Table 5.1-1 tabulated individual cor-

• rections to the fission dosimeter measurements would be very informa-tive, as would a discussion in Section 5 of the procedure used in cal-culating them, including a listing of the photofission cross sections used.

I think too much emphasis is being attached to the adj ustments calcu-lated by FERRET on the (M/C) ratios before and af ter adj ustment. Vir-tually any least-squares adj uscment procedure that includes uncertain-ties will yield adjusted calculated values very close to the measure-ments simply because the measurements have considerably smaller uncer-tainties, although the measurements are adjusted as well. This is a necessary but not sufficient condition for testing the procedure. A 6

. =.

chi-square per degree of freedom test is a more meaningful one.

The data in Section 6 taken in the cavity at axial positions near the bottom of the core with the chains obviously cannot be calculated with DORT R9. Indeed, none of the axial profiles can be calculated with DORT R9, and I question the reasoning behind including any comparisons of axial measutements with midplane calculations.

Is there a typo in Table 6.3-1 for capsule 5 and the Ni-58 reaction rate? It seems out of line with all the other Ni-58 entries. There also seems to be an inconsistency among the values for the fission reaction between capsules R and U, which are only 2 degrees apart in azimuth. compared with the top four reactions. I wonder if this is in-dicative of inaccuracies in the calculated corrections for these two reactions. Even the azimuthal profile of the U-238 fission reaction behaves strangely compared with the top four reactions. Does the un-corrected data have similar behavior?

For reasons already given, some of the data in Table 7.1-1 are not only erroneously labeled as " measured" but I seriously question the worth of these data. To my mind, the only worthwhile measured and cal-culated data comparisons appear in Table 7.2-1, from which in-vessel (M/C) values average about 0.92 and ex-vessel values 0.87. I also ob-ject to the use of the words " bias factor", used throughout, since it implies per se a final correction factor without undergoing any furth-er "best estimate" adjustment procedure. It is simply a " measured to calculated ratio". As I understand the nomenclature, the "best estim-ate exposures are the result of a " massaged" series of bias factors.

In Section 8.2, many important sources of uncertainty in the calcula-ted fluxes are not mentioned. They include the absolute source spectra leaking the core and method and modeling approximations such as 2-D description of a 3-D geometry (i.e.,

effects of a finite axial core),

steel densities, azimuthal variations in the PV inner radius, composi-tion and water content of the biological shield for the cavity calcu-lations, and inaccuracies in the BUGLE-93 iron cross sections. For the fission dosimeters, negligible uncertainties arise from gamma-ray transport but there are considerable uncertainties in the photofission cross sections as well as in thermal fluxes calculated with two groups without upscatter. These fluxes are the sources for the secondary gamma-rays produced from neutron capture and probably dominate the

,high energy flux where most of the photofission reactions occur. The calculated cavity flux uncertainties in Table 8.2-1 are probably low by a factor of about two.

The following pages offer my comments on the RAI's from USNRC and the responses submitted by Consumers Energy Company. Generally speaking, I found the RAI's meaningful and to the point, and often were the same questions and requests for clarifications which arose in my mind as well.

In CEC's response to NRC issue 1, it is stated that the uncertainty associated with calculations alone is typically 15%. For surveillance capsules and PV inner radius this is reasonable, but not for the cavity fluxes. Here, the iron inelastic cross sections have uncertain-7

c ties such that, even with BUGLE-93, produce significant flux uncer-tainties in the neighborhood of 15% all by itself. Adding in the con-tributions from other sources increase the overall uncertainties to around 25%. The statement is also made that independent evaluations based on both the 20 plant Siemens-KWU data base and the 21 plant Westinghouse data base tend to result in overpredictions of the calcu-lated fluence. I submit that this may be due to consistent overpredic-tive methods used in the core leakage (at least for Westinghouse), and the analysis based primarily on surveillance capsule data. However, even for in-vessel locations, calculations using BUGLE-93 still tend to underpredict both the~PCA and HBR-2 benchmarks by up to 10%. For cavity dosimetry, if they employ only DORT-RO calculations the Germans would be overlooking effects of the finite axial core j ust like West-inghouse. Using any cross section library that predates ENDF/B-VI data will also severely underpredict cavity fluxes by around 35%.

In CEC's response to NRC issue 2, I question the meaning of the word "small" in "a small variation in the calculated vs. ' measured' energy distribution also exists". For the cavity fluxes especially, there can be significant changes between the calculated and adj usted fluxes caused by adjustments in the inelastic iron cross sections. For exam-ple, a change in the iron inelastic cross section of 6% changed the cavity group fluxes in HBR-2 by 25%.

In CEC's response to NRC issue 4, I completely agree with the list of possible reasons for the mismatch between calculated and measured dos-imeter responses. An identification of which sources of uncertainty are most responsible f or the adjustments in the calculated group fluxes is part of any def ensible adj ustment procedure. I am also in agreement with the final paragraph in the response, which states that " specific plant measurements represent a practical means to quan-tify the net overall bias in the calculation". But I add: it must con-sider the uncertainties in the measurement as well. It is possible to separate out effects arising from inaccurate cross sections and other generic data by a simultaneous adjustment of simple, well-defined benchmarks and a power reactor. The benchmarks in this case would con-sist of such experiments as the PCA and PSF measurements at ORNL.

This leaves the data unique to the power reactor such as core source and the physical plant specific data to be adjusted without so much interference from the generic data. Worded another way, the generic data is adjusted primarily by the benchmarks since their sensitivities to the data are high and relatively free of other uncertainties, which are then corroborated by the power reactor adjustments, thus leaving more freedom for the plant specific data to be adjusted on their own merits. The general discussion following NRC issue 4 has lots of good things in it, and clears up the relationship between "Best Estimate" and the (M/C) ratios. I was also delighted that there is no mention of the term " Bias Factors" anywhere in this discussion.

In reading the more detailed presentation of NRC issue 1 I am now even more concerned about the advisability of invoking the results of the Siemens-KWU reactor analyses to support Palisades' contention that f

overprediction is common. Do the Germans use a calculational methodol-ogy identical to Westinghouse? Are their reactors virtually identical 8

in design to those of Westinghouse? Even if the answer is in the affirmative, aren't all reactors subject to their own plant specific sources of uncertainty, even those of Westinghouse which have the same calculational met'hodology?

In the more detailed discussion of NRC issue 2, the presentation of the (M/C) comparisons in Tables 2-1 through 2-6 may be hiding some im-portant information. Since the data are not separated for a given data base into in-vessel and ex-vessel locations, they tend to obscure any effects of PV transmission inaccuracies. It would be monumentally better to so divide them along with the number of entries at each loc-ation. In addition, some indication of the transport methodology and cross-section library as well as reference to dosimetry cross sections used for the St. Laurent and HBR-2 calculations would be helpful in interpreting thse tables.

In the more detailed presentation of NRC issue 3, the transport meth-odology and cross sections including dosimetry that were used in obtaining the (M/C) values for Siemens-KWU in Table 3-1 should be iden.

tified. Partitioning into in-vessel and ex-vessel locations as men-tioned above for the tables in NRC issue 2 would be informative in this as well as all the remaining tables in NRC issue 3. It must be significant that Palisades ranks consistently above all but one of the other 20 Westinghouse reactors in the overprediction of the measure-ments. This cannot be a random occurrence. Perhaps this is connected with the absence of a thermal shield, but how?

In the more detailed presentation of NRC issue 4, my attention was centered on Table 4-1, which presented a summary of potential sources of calculational bias. I was gratified to see 8% on the accuracy of the peripheral pin powers and another 8% on the cavity fluxes due to inaccuracies in the transport cross sections. The source uncertainty agrees very well with what is necessary to account for the (M/C) dis-crepancies at the in-vessel locations (assuming it is too bigh) and an additional iron inelastic cross section uncertainty of 2% would re-sult in about an 8% uncertainty in the cavity fluxes (the sensitivity of the cavity fluxes to changes in the cross sections is about -4),de-creasing to a 4% effect for mid-PV fluxes and to essentially no effect at the PV inner radius. (Studies have shown that the flux sensitivi-ties to modest changes in the iron inelestic cross section are linear with thickness of iron penetration.)

In the response to item 1.2 in attachment 1 of the letter dated June 21, 1996 from Consumers Power to USNRC, the first table shows an esti-mated 15-16% increase in the calculated ex-vessel capsule fluxes in going from ENDF/B-IV to ENDF/B-VI. My experience indicated about a 25% increase in the region 3-8Mev and somewhat less than that below 3Mev. Another question I have about this response is how does explicit modeling of the ex-vessel dosimetry result in up to a 23% increase in the fluence? The dosimeter containers were so thin that they weren't modeled in the calculation at all. Have the cavity dosimetry calcula-tions been corrected for these 3-23% effects? This is terribly confus-9

ing. At what radius are the percent changes at the bottom of page 2 applicable? If it is that of the eccelerated capsule, I am sur-prised that an increase in the PV inner radius of 0.12in. has a 3% effect there.

In the discussion concerning the use of the C jin the equation re-lating the reaction rates to the activities, it is stated that for single cycles C j. is taken to be 1.0.

Does this mean the changes in the core spatial power distribution during a cycle are assumed to be small and disregarded? Can't these be.significant as well, or is the midcycle distribution considered to be an adequate approxima-tion?

The response to item 2.1 does not address in sufficient detail the calculation of the thermal flux and its accuracy or the listing of the photofission cross sections of Verbinski and their estimated accuracy.

The response to item 2.3, as shown in the tables on pp. 13-17, is very good. It would be even better if two additions were made:

one, corrections from U-235 contamination and Pu-239 build-in list-ed separately and two, estimates of the uncertainties in all the entries presented. These uncertainties impact those of the correct-ed fission dosimeter measured reaction rates.

In looking at the equation relating the reaction rate to the spec-ific activity again in subsection 3.1.2.3, since the reaction rates are normalized to a power of Pref but the dosimeters were only re-sponding to power levels some 2% less because of the venturi prob-lem, shouldn't all the measured reaction rates be increased uniform.

ly by 2%? Has this factor been incorporated into the final results?

Why wasn't this mentioned in subsection 2.37 l

l j

The uncertainties in the reaction rates as deduced from the activ-ity measurements shown in subsection 3.1.2.5 seem very reasonable.

The only caveat I have stems from the (almost necessary) cavalier treatment of uncertainties in the corrections for photofission. I'm sure the cross sections have large uncertainties and the uncertain-ties in the gamma-ray source spatial distribution arising from thermal-neutron capture may also be large. A combined uncertainty of a factor of two or even higher doesn't seem at all unreasonable.

'I do think the 25% uncertainty assumed for the photofission correc-tion is unduly optimistic.

As I have previously pointed out, the various components of the group flux uncertainties which are due to specific causes such as iron cross sections, core leakage, transport method approximations, tolerances, etc., must be established as independent parameters in any credible adj ustment procedure, because it is the adj ustments in these parameters which give rise to adjustments in the fluxes at not only the dosimetry locations but at others such as in the PV where no measurements can be made. These " translations" involve the use of PV flux sensitivities and permit one to obtain results of the adj ustment procedure anywhere, including reduced adjusted uncertainties as well. From the description of the input to FERRET 10

given in subsection 3.1.2.7, or the other hand, it appears several important ingredients are missing.

As I understand FERRET, it treats all the dosimetry measurements the same, i.e., the in-vessel and ex-vessel (M/C) values are put on an equal basis, even'though there is a larger uncertainty in the calculated cavity fluxes primarily because of inaccuracies in the iron cross sections and method approximations. The adjustment procedure then seeks a best solution to the overdetermined system of all the (M/C) values, the dosimeter cross sections, and the cal-culated group fluxes, subj ec t to their covariances (standard devia-tions and correlations). The solution is found and the adjustments indicate the measurements, by virtue of their smaller uncertainties, do not change much, nor do the dosimeter cross sections, for the same reason. The calculated group fluxes do change, however, and yield new values of the (M/C) all close to unity and which give rise a single correction factor, BE/C, which is used to multiply the calculated fluxes at all locations. It is this latter that ren-ders FERRET deficient, because it fails to take into consideration the changes in the basic data required to effect the necessary changes in the calculated fluxes, i.e.,

all those quantities that contribute appreciably to the flux adjustments. If the adjustments in these individual parameters were identified and combined with the flux sensitivities appropriate to the parameter for a given PV location, then the " translation" problem is solved. In this way, l

cavity dosimetry has the potential to be j ust as effective in pinning down PV fluences as surveillance dosimetry. The trick is in the sensitivities,which can be readily calculated, and keep-ing track of the individual parameter adjustments. FERRET treats all the sources of flux uncertainty the same in the sense they all have flux sensitivities of unity for all groups and all locations.

Thus it doesn't matter which of the parameters is adj usted, it aff-ects all the PV fluxes in the same relative way. The spectrum ad-justed by the dosimetry comparisons is compared to the unadj usted spectrum at these points, and the ratio becomes the "best estimate" divided by the original calculation, both spectra summed over the same energy range such as above 1Mev. This ratio becomes a renor-malizing factor to " correct" the fluxes at all locations. As an ex-ample of a situation which clearly shows up errors in this code, suppose the (M/C) values in the cavity averaged 1.35 and the sur-veillance values averaged 0.95. How would FERRET combine the two dosimetry sets to arrive at PV flux adjustments without knowing l

'that the cause of the cavity underprediction was very likely due to inaccuracies in the iron cross sections and that these would have a much smaller effect on the PV inner radius fluxes where neutrons j

have only had to penetrate the small thicknesses of the shroud, core barrel, and PV cladding?

I believe the use of a correct adjustment procedure would render as probably no longer true the statement made in Section 3.3, which is that changes in the calculation that are characterized as generic do not impact the best estimate fluence calculation for the Pali-sades reactor vessel because if these changes were not implemented, the calculations and the bias factor would be affected in equal and opposite amounts, leaving the best estimate unchanged. This is no M

longer true if the original calculations are modified by something more than a simple renormalization factor, because now every group flux at every location throughout the entire calculation geometry will have a different BE/C, i.e.,

there is no longer one " bias fac-tor" but many.

In attachment 1 to the letter dated September 9, 1996, from CEC to NRC, the response to item 2.1 is in error. There is no reason why the (M/C) average biases for in-vessel capsules and cavity cap-sules should be the same, generally speaking. The reasons they should in general be different have already been discussed in this letter. The only exception is if the only source of uncertainty is due to core source magnitude, in-which case the calculated fluxes would be changed by a constant factor. It is just fortuitous that for Palisades the two averages are so close to each other (0.87 and 0.92), for as we have seen, the causes of the mismatches at the two locations are completely different.

In the response to item 2.7, no mention was made of including cor-relations as well as the standard deviations in describing the dos-imetry measurements. I do not understand the first paragraph on page 13.

In the response to item 3.10, it is stated that a post-adjustment uncertainty was added to the uncertainty of the adjusted best es-timate. This should be avoided at all costs, because an uncertainty has an effect on the least-square adj ustments, and should be con-sidered in'the procedure.

In attachment 1 to the letter dated September 19, 1996, again from CEC to NRC, the response to item 3.1 contains Table 3.1-1 in which three of the adj usted reaction rates fall outside both the measured and calculated values. Is this a violation of any mathematical law?

My answer is probably not, because there are uncertainties in both the measured and calculated values which probably cover a range wide enough to encompass the adj usted value. It would be interest-ing to view these standard deviations, however. In Table 3.1-2, where are the fluxes located? The calculated flut uncertainties are quite high; they should be larger than those of either the measure-ments or the dosimtry cross sections, but 42.4% all the way down to 9.12 Kev seems to indicate a complete lack of confidence in the cal-chlation of the in-vessel fluxes in particular. More attention must be devoted to obtaining reasonable estimates of these uncertain-ties. In Table 3.1-3, something is terribly wrong. The uncertain-ties of the adjusted reaction rates in six of the eight dosimeters are larger than those of the measurements alone. It is a fundament-

)

al rule of information theory that adding more information such as, in this case, the calculated reaction rates, no matter how large the uncertainty, must result in added knowledge with a consequent decrease in the combined uncertainty. Comparing the entries in Tables 3.1-6 and 3.1-7 raises another question: Why are the adjust-ments in the Ni-58 cross section so much smaller (last column) than for Fe-54 even though the uncertainties for the Ni-58 cross section are considerably larger? In Table 3.1-9, in the region between 0.4 12

and 1.0Mev,Np-237 is the only dosimeter with an appreciable response, yet there is no cross section adj ustment. Similar inconsistancies are evident in Table.3.1-10 as well. At group 38, even with a unique sensitivity to this flux, the cross section doesn't change, although in this case the adjur,tment procedure may not have included this low energy reaction. In Table 3.1-11, the adjustment in the calculated flux above 1Mev is two-and-a-half times its uncertainty, which is a very unlikely occurre1ce,but not forbidden. Another inconsistency involves the 8% unc.ertainty in the calculated cavity fluxes above 1Mer and the 42.4% uncertainty in the calculated in-vessel fluxes shown earlier in Tables 3.1-2 and now in Table 3.1-12 as well. There occurs another instance of increases in the uncertainties after ad-justment in Table 3.1-13, first detected in Table 3.1-3. Table 3.1-17 shows there were absolutely no adjustments performed in the Ni-58 cross section for the 16 degree Cycle 9 dosimetry capsule, further evidence that the adjusted Ni-58 data are erronious. The fractional changes in the Np-237 fission cross section as a result of adj ustment shown in Table 3.1-19 for the 16 degree Cycle 9 cavity dosimeter look more reasonable than those in Table 3.1-9 for the Cycle 9 in-vessel capsule, but still suffer from an unexplainable break at group 24. Again, Table 3.1-10 duplicates the strange behav-ior of the low energy Cobalt reaction observed earlier in Table 3.1-10.

I found the RAI's 3.4, 3.5, 3.6, and 3.9 to be interesting requests.

In item 3.9 I feel that the rapid convergence to a solution that is demonstrated by iteration is a necessary but not sufficient con-dition to validate an adjustment code. As previously mentioned, the chi-squared per degree of freedom test, which is a measure of the average adj ustment of all the parameters entering into the procedure in units of standard' deviation, is a more reliable test. A value in the close vicinity of unity indicates a very consistent set of data.

In attachment 1 of the letter dated October 1, 1996 from CEC to NRC, the response to item 1.4 elicits the following comment. Amomg the sources of uncertainty that contribute to flux uncertainties that fall under the heading of method approximations are such effects as the neglect of the finite axial source previously discussed. These approximations may ^ either be corrected'by other subsidiary calcula-tions or left uncorrected. In either case, a bias factor needs to be introduced as another source of uncertainty to accommodate either the uncertainties in the correction factor or the uncertainties in the uncorrected calculation. Of the two, the results from the sub-sidiary calculations are to be preferred, since otherwise the magni-tude of the effect must be estimated beforehand with uncertainties.

A method based on flux synthesis, previously alluded to in this letter, should offer a reasonable correction factor, and after using it to modify the original calculations, the uncertainties in this bias factor can be entered into the adj ustment procedure as rela-tively refined estimates of a few percent.

The response to item 2.7 states the reason for the use of a log nor-mal based least-squares procedure in FERRET as being the constraint 13

of dositivity it places on all fluxes. This is correct, but my feeling is, if you need this requirement to effect a physically permissable adjustment, the input data including uncertainties must be grossly inconsistent. The fractional adjustments should be small enough that the condition of linearity in the flux sensitivities is preserved, and this has found to be.the case when the disagreement between cal-

~

culation and measurement is within 30% or so. Under these conditions a normal least-squares procedure is preferable.

In the response to item 2.10, I think that it might be too early to draw any definitive conclusions about the " bias" that may exist be-tween the high threshold dosimetry (M/C) comparisons and the others with lower thresholds. It is still very possible that there are in-accuracies in the iron cross sections throughout the erergy region below 3Mev, where the iron cross section structures for both elasti filled with many' peaks and valleys, un-and inelastic scattering are like the region above 3Mev. A possible deduction is, therefore, that this apparent " bias" is real and the cause lies in the other four lower threshold dosimeters because of transport inaccuracies,and not in the two highest threshold dosimeters. No such spectral dependence was observed in the HBR-2 analysis which used the same dosimeters, f

however. Finally, the response also includes the statement that the (M/C) values for the Fe/Ni/U/Np group of dosimeters are more heavily weighted than the Cu/Ti pair. If by " weighted" is meant in the ad-justment procedure, other than the fact there are four values com-pared to two, this should not be done because energy dependence is important in the slowing down of neutrons with the higher energy fluxes feeding the lower energy fluxes.

The following are my comments on the letter dated December 20, 1996 from John Hannon to Thomas Bordine and the two enclosures. I fully agree with the NRC decision to accept the 8% PV fluence reduction due to updated plant specific information. I also fully agree with the decision not to approve at this time (i.e.,

1996) the remaining 17%

based on the analysis of the dosimetry data as provided by CPC/W. My reasons for the latter are basically the same as those given by BNL and NRC, which are that they are based on the results of an adjust-ment procedure which I strongly feel is deficient. I have attempted to present the faulty logic which I feel exists in FERRET for use in the present application. I have also cited specific results that were output from the code which strongly suggest their questionable accur-acy. I don't believe Fred Schmittroth, the author of FERRET, ever had the complications of the present application in mind when he wrote i t, I think he intended to use FERRET as a means for reconciling the over-determined problem of a series of dosimetry calculated and measured responses at a SINGLE LOCATION. Further, as I have also tried to make clear, uncertainties in the calculated fluxes are not adequately de-tailed, and sensitivities which correlate the PV fluxes with the fluxes at the dosimeter locations are nct even considered. I would be the first to tell you that I have not taken the time to understand the mathematics in FERRET, and I an only guessing at what it does.

But I have had a great deal of experierce in both reactor fluence calculations as well as assisting in the development of the LEPRICON adjustment code, which performs all the steps alluded to in this letter. LEPRICON has not been actively promoted in the past several years, but it is available in RSIC at ORNL.

nA

l

(

I have one comment on Section 1.0 in the Safety Evaluation Report.

The alternative explanations suggested for the differences observed be-tween the (M/C) values of the highest threshold dosimeters and the low-er threshold ones are neither very likely since a 14% difference cannot l

logically be made up from 6% inaccuracies in both the Fe-54 and Ni-58 cross sections and another 6% or so in both the Fe-54 and Ni-58 measurements. A more likely cause of these spectral effects lies in inaccuracies in the iron cross sections in the 1-3Mev region as has already. been mentioned in this letter. It is also possible that the changes in the spectrum due to effects from a finite axial core could i

be significant.

In Section 2.0 I would like to see a detailed validation performed of the basic peripheral assembly powers since it was not addressed by BNL.

l If it was about 5% high in the calculations, then my cursory analysis l

would indicate this to be a valid reason for a reduction in the fluence

at the PV inner radius with a corresponding 5% increase in the time re-L quired to reach the PTS screening criterion.

l In Section 5.0 I was absolutely amazed that the two group with no up-scattering thermal treatment in BUGLE-93 agrees so remarkably well with the 42-group with upscattering treatment of the MATXS12 library that I

l BNL used, with different photofission cross sections, in the correction factors to the fission dosimeters. Perhaps my estimate of a factor of two or even more uncertainty in these correction factors was too pessi-f mistic.

l In Section 6.3, I again make the point that I agree with the conclusion t 'ha t for the present application the results from the FERRET code are not credible and do not represent reality, i

In conclusion, I would like to apologize in retrospect for repeating certain comments throughout this report, but the fact is that they i

needed reinforcement at times in response to queries posed by the NRC as well as to answers supplied by CEC. My final comments involve l

the status of the request by' CEC to NRC for acceptance of a further 17%

reduction in fluence which would extend the life of Palisades until the year 2011. On the basis of what I have presented in this report, the fact that the cavity dosimetry results were overcalculated by 12% has little influence on the calculated fluence at the PV inner radius if j

the causes of the overcalculation are due to inaccuracies in the calcu-lation that do not affect to any significant degree the in-vessel loca-tions. Evidence that I have detailed suggests such is the case. Of the two probable sources of overprediction in the cavity - neglect of finite axial core effects and source normalization, only the latter has any strong effect at the PV inner radius, estimated to be a flat 5%

everywhere, and would be a legitimate reason for consideration if it can be validated by core instrumentation. The postulate that the iron cross sections are still responsible for a 10% underprediction of the cavity fluxes, if true, also would not affect the calculated PV inner i

radius fluxes. Cavity dosimetry, to be effect1ve, must rely on very accurate transport calculations in order for them to be useful.

k Richard Maerker 15 l

l

)

i 1

ENCLOSURE 4 CONSUMERS ENERGY COMPANY PALISADES PLANT

{

DOCKET 50-255 l

l Palisades Reactor Vessel Fluence Resume & Independent Review Completed by W. N. McElroy Consultants and Technology Services I