ML20198P461

Submits Review of Calculations Provided for Licensing of McGuire Nuclear Station Spent Fuel Racks & Util Nuclear Licensing Rept for High Density Spent Fuel Racks for Oyster Creek Nuclear Generating Station

ML20198P461
Person / Time
Site:	Mcguire, Oyster Creek, McGuire, 05000000
Issue date:	07/20/1984
From:	Belytschko T AFFILIATION NOT ASSIGNED
To:	Kim B NRC
Shared Package
ML20151H203	List: ML20151H200 ML20151H205 ML20151H295 ML20151H817 ML20151H829 ML20151H860 ML20151H871 ML20151H975 ML20151H984 ML20151K379 ML20151K471 ML20198N913 ML20198N918 ML20198N951 ML20198N993 ML20198P175 ML20198P197 ML20198P243 ML20198P290 ML20198P353 ML20198P407 ML20198P461 ML20198P551 ML20198P560 ML20198P632 ML20198P644 ML20198P647 ML20198P673 ML20198P772 ML20198P803 ML20198P818 ML20198P858 ML20198P868
References
FOIA-86-26 TAC-53531, TAC-53532, NUDOCS 8606060315
Download: ML20198P461 (3)

Text

i, %,

\\

~

g s

)

s 3

S l

/

fa 7bi/

TED BELYTSCHKO, !NC gu' 2 18 LONGMEADOW ROAD WINNETKA, IL 60093 (312) 446-7029 / 492-7270 July 20, 1984 Mr. Bung So Kim U.S. Nuclear Regulatory Commission Philips Bldg.

7920 Norfolk Ave.

Bethesda, Maryland 20814

Dear Dr. Kim:

The purpose of this letter is to review the calculations provided for the licensing of the Duke Power Company McGuire Nuclear Station Units 1 and 2 spent-fuel racks and the GPU Nuclear Licencing Report for high density spent-fuel-racks for Oyster Creek Nuclear Generating Station (NRC docket number 50-219). $ comments will be limited primarily to the computational procedures employed in the model, except for the several items which I was specifically asked to consider, I will not comment on aspects such as the modelling or the specific model parameters used since the review of these features would req; ire far more time than I was asked to devote to the project.

I would also like to point out that because the licensing reports did not include many details on the actual methods used, I have relied heavily in my evaluations on oral reports presented by Mrs. Vu N. Con, Vincent K. Luk and R.C. Herrick who have been briefed by the people who performed the analyses. These present-ations were made to me at the Franklin Institute in Philadelphia, Wednesday July 18th,1984, and I will henceforth call this the Franklin presentation.

In the McGuire procedure, a combination of a two-dimensional nonlinear analysis and a three-dimensional linear model is used in order to establish stress levels. An implicit time integration procedure is used for the nonlinear model.

It has been stated that a large variety of time steps have i

been tested in the nonlinear analysis and the time step chosen for the final results falls in a large window of time steps for which the results are relatively independent of the time step; this is a good indication that the j

solution has been probably converged.

It is thus to be expected that the maximum displacements predicted by the analysis are relatively free of numerical errors and can constitute a basis for a reliable licensing calculation.

The only area in which the M:Guire Report is subject to question is in the procedures to factor the two-dimensional nonlinear results into a three-8606060315 060319 PDR FOIA PATTERSOB6-26 PDR 4-102_

\\.,.

y 2

4 B.S. Kim Page 2 i

i dimensional linear model for the stress analysis. This procedure is described very tersely on pages 2.3-1 to 2.3-2 in the report. On these pages, it is stated that "the loads used in the linear seismic analysis are corrected by I

load correction factors obtained from the nonlinear analysis." The oral 1

reports I reviewed on the applications of this load correction procedure 4

j indicate that the results of the linear model are scaled up by the ratio of the moments and shears generated in the two-dimensional nonlinear analysis as compared to the linear three-dimensional analysis, and that the two horizontal and the vertical components are then combined by the SRSS method. Although intuitively this procedure appears to me to be conservative, I do have some j

i reservations:

i

)

(1) I have never before seen this method applied to obtaining relation-ship between two dimensional nonlinear and three dimensional linear analyses, i

so the method is unfamiliar to me and I do not know its background.

I (2) Since the loads appear to be scaled by the moments and shears on the bottom surface of the fuel rack, it is not clear how the nonlinear effects of i

i rocking which result primarily,in axial loads and moments in the horizontal members are compensated for in going to the three dimensional linear model.

It would appear that this deficiency could be corrected by several alternatives:

j (1) use of a three dimensional nonlinear analysis; i

(2) citation of previous studies and applications of this load correction factor method; 4

(3) simplified studies to show that the method is conservative, or can be made conservative through the introduction of additional correction factors.

]

The Oyster Creek analysis was performed through the use of a three dimen-sional model which was integrated by an explicit time integration technique.

Again, the major difficulties with the procedure were described in the j

Franklin presentation and involved the large variation of results obtained l

with small changes in the time step.

It is well known that explicit methods 1

are sensitive to the choice of time step and that when the stable time step is exceeded by even a small amount, the resulting calculation will be unreal-istically large, which many computational engineers describe as the solution

" blowing up".

Numerical instabilities are usually quite obvious because the displacements which are predicted are ridiculously large, but in some non-linear calculations the instability may be arrested by a dissipative mechanism such as friction, which can then result in a computation of unrealistically high values. These unrealistic values are not readily detectable by an analyst. Unless an energy balance check is included in the computer program. This was noted by me back in 1974 and I have included a copy of the relevant paper.

i i

I e

,_ -.. -,.. _,,.. ~. _. -. - _ -., -. - - - - - -. _, _., -., _

\\_

(

1 B.S. Kim Page 3 Generally, for a large range of time steps below the stability limit, the solution should be quite independent of the time steps. However, if the time step is very small compared to the period of an element in the system, round-off errors can again result in the failure of the explicit method.

In most engineering models which I have dealt with, the window of accept-able time steps, however, involves at least on order of 10 getween

• • 9 3

and the smallest time step and more generally as much as 10 or 10.

In the calculationfortheOysterCreekfacglityaconvergedsolutionwasobtained for only a single time step, 2 x 10- sec; calculations exhibited an instability for 3 x 10-b and failed to produce a solution at 1.5 x 10-5 sec.

This very narrow window of possible time steps may be due to the very high 1

stiffness of the springs used to model the feet of the rack. Nevertheles present,itcannotbeestablishedwhethertheresultsobtainedat2x10g.at sec are reliable because ig is not clear that the factors which completely prevent solutions at 1.5 x 10- sec are also not operative at the slightly larger time step.

In any case, before reliability can be ascribed to these calculations, it j

is necessary to show that for a reasonable range of time steps, such as a

,)

factor of 2, the results are relatively insensitive to the time step.

I would also like to point out, that the reliability of any explicit computation would be enhanced considerably by an energy balance check which would demonstrate that the computational model has not violated the conservation of energy.

If you have any questions, please feel free to call me.

J Sincerel,

/

/

l h

Ted Belyts ko i

1 TB:sg

\\

\\

cc: Mr. C. Herrick

.. _ _