Part 21 Report - 1997-790

 
 

ACCESSION #: 9710080070



                 Holtec Center, 555 Lincoln Drive West, Marlton, NJ 08053



                                                 Telephone (609) 797-0900

HOLTEC                                                 Fax (609) 797-0909

INTERNATIONAL



June 27, 1997



Mr.  Thomas Greene, NRR/PECB

U.S.  Nuclear Regulatory Commission

Mail Stop: 0-11E4

Washington, DC 20555-0001



Reference: Accession 9601190322



Subject:  Request for a Technical Review of a Draft Information Notice

          Regarding Erroneous Results from ANSYS 5.0A and ANSYS 5.1

          Computer Codes



Dear Mr.  Greene:



On October 16, 1995, Holtec International had issued a 10CFR21 notice to

Region I of the USNRC advising the Commission of our company's discovery

of a serious deficiency in the general purpose finite element Code ANSYS.



Our analysts found that ANSYS produced non-unique results (different

results on different computers) when a dynamic model of a free-standing

submerged fuel rack for a DOE facility was subjected to 3-D seismic

excitations.  Inasmuch as free-standing spent fuel racks in spent fuel

pools are stored at numerous commercial and military installations, this

discovery of numerical instability in ANSYS warranted that we inform the

Commission immediately.



ANSYS is part of Holtec's library of QA validated codes, but we have

administratively prohibited the use of this Code over the past ten years

in dynamic analysis of free-standing underwater structures.  Our initial

reason was the lack of suitable elements in ANSYS to model fluid coupling

effects in multi-body problems (such as array of fuel racks in a pool).

All versions of the Code to date have continued to suffer from the above

deficiency with respect to modeling of fluid coupling in an array of

racks.  The 1995 analysis was conducted at the behest of the DOE, which

led to the discovery of the numerical discrepancy in solutions obtained

on different computers.



                 Holtec Center, 555 Lincoln Drive West, Marlton, NJ 08053



                                                 Telephone (609) 797-0900

HOLTEC                                                 Fax (609) 797-0909

INTERNATIONAL



Mr.  Thomas Greene, NRR/PECB

U.S.  Nuclear Regulatory Commission

June 27, 1997

Page 2



As the attached Holtec internal memo states, verbal interactions between

Holtec and NRC personnel may have obfuscated the true nature of the

deficiency in ANSYS.  Therefore, I request that the Commision review the

information provided by our Dr. Soler in the attached memo, and act on

our Part 21 notification accordingly.  We have no knowledge of whether

the problem still exists in more recent updates of the ANSYS Code.



ANSYS is a robust finite element Code which has served as a reliable

workhorse in our day-to-day structural analyses.  However, we continue to

treat this Code as unsuitable for simulating complex nonlinear transient

problems (such as dynamic modeling of spent fuel racks) which may be of

safety consequence to the nuclear industry.



By our notification of October 16, 1995, our company has discharged its

obligations per 10CFR21.  However, Holtec QA feels an obligation to

provide to your attention the contents of the attached inter-office memo.

We would be pleased to provide additional technical information if

required by the Commission.



Sincerely,



Vikram Gupta

Quality Assurance Manager

VG:nlm



cc:  Mr.  Gautam Bagchi, Structural/Geophysical Branch

cc:  Chief, Region I, King of Prussia, Pennsylvania



Attachment: Inter-Office Memo from A.I. Soler to V. Gupta



                         INTEROFFICE MEMORANDUM



TO:       Vik Gupta, Q.A.  Manager



FROM:     Alan I. Soler, V.P. of Engineering, Holtec International



DATE:     June 23, 1997



SUBJECT:  Problems with use of ANSYS in Fuel Rack Analyses



On Wednesday, June 18, 1997, a meeting was held concerning spent fuel

rack licensing for the TVA, Watts Bar plant.  During the course of that

meeting, I had occasion to reference the 10CFR21 issue on ANSYS 1_/

raised by Holtec (which eventually resulted in a phone conversation

between Holtec and the NRC subsequent to ANSYS Inc. presenting some

additional material in June, 1996).  The Branch Chief of Structures and

Geosciences, Dr.  Bagchi, questioned me on the subject and quite

emphatically asked why Holtec withdrew the 10CFR21 notification.  I

replied to him that to my knowledge and recollection Holtec did no such

thing, and that while I did acknowledge over the phone and in a

subsequent fax transmission that had ANSYS provided to me the information

that they provided to the NRC, I might have been able to reconsider the

issue, that verbal and taxed statement in no way implied a "withdrawal of

the notification".  It is not even clear to me that Holtec is permitted

to withdraw such notification and even if we are, such notification would

come from the QA manager at Holtec.  At the time, It was my understanding

that NRC was going to make a decision internally concerning the issuance

of the notification, and it is my recollection of the phone conversation

that Dr. Kim (structural branch) thought it should be issued even if only

to warn the industry that despite all of the disclaimers in the ANSYS

manual, non-linearities are not easy things to deal with and that

problems may occur with even the most experienced users.



There appears to be the incorrect perception at the NRC that we caused

NRC to expend resources to evaluate a potential problem and then

retracted the query at the last minute.  This perception should be

dispelled.



Subsequent to the June 18, 1997 meeting concerning Watts Bar (which does

not involve any use by Holtec of ANSYS for non-linear dynamic analyses),

I have given more thought to the 10CFR21 notice issue observations: I

have the following:



_________________________

1_/  Holtec Part 21 letter dated October 16, 1995 to Region 1, USNRC.

_________________________



1.   ANSYS is the general purpose finite element code that has been used

     in the past and continues to be used at Holtec for a wide variety of

     linear and non-linear analyses.  In accordance with our own internal

     QA requirements, we independently confirm that the example problems

     provided by ANSYS are reproducible on the PC computer platforms in

     use at Holtec.  Each version of ANSYS is subject to a comprehensive

     Q. A. validation process before it is accepted for use within the

     company's Q.A. system.



2.   As a result, ANSYS is qualified for use in a wide variety of

     problems, except the specific class of nonlinear problems involving

     underwater response of free standing structures.  Holtec, with a

     single exception of a DOE project, has not used ANSYS to simulate

     the non-linear, time dependent behavior of an array of spent fuel

     racks in a spent fuel pool subject to a 3-D seismic event.  Over 20

     years ago, we developed and validated our own computer code which is

     not a general purpose code, but is directly focused on the

     particular aspects of spent fuel rack analysis which are not

     amenable for modeling by a general purpose code.  The special

     treatment of near field and far field fluid coupling of a generally

     rectangular array of closely spaced bodies, the fact that the

     physical problem involves a multitude of free standing bodies which

     may make and break contact with adjacent bodies, the fact that

     frictional effects must be considered at some of the same locations,

     and the fact that the solution requires a time history analysis with

     appropriate time stepping algorithms taken in aggregate make this

     simulation of the behavior of an array of spent fuel racks a complex

     problem.  ANSYS, in some of their responses, admit to the

     inabilities of the code to analyses such complex problems.  Further,

     in my opinion, the existing fluid element in ANSYS, developed for

     two concentric cylindrical bodies with a small fluid filled annulus

     between the bodies, is not capable of correctly modeling the near

     and far field effects of an array of adjacent rectangular bodies.

     As I noted to Dr.  Bagchi during the Watts Bar June 18,1997 meeting,

     Holtec: has used ANSYS in only a single spent fuel rack application

     to the Department of Energy at the specific request of DOE.  It was

     this single application of ANSYS to the spent fuel rack simulation

     that led Holtec to issue the 10CFR21 notification which raised an

     issue of multiple solutions occurring on different computer

     platforms.



3.   The complete record of the various submittals to ANSYS by Holtec and

     the responses from ANSYS to Holtec was given to the NRC by Holtec as

     part of the 10CFR21 package submittal.  It is not necessary here to

     rehash the details of those documents here.  However, among other

     items, the responses by ANSYS acknowledged that they felt that the

     problem was highly complex, involved severe non-linearities, and

     suggested that additional damping would possibly eliminate the

     multiple solutions.  The lack of any more specific root cause

     evaluation on ANSYS's part left me with no recourse save to comply

     with federal regulations and issue the 10CFR notification in a

     timely manner.



4.   Subsequent to the NRC issuance of a draft notice for comments to

     both Holtec and to ANSYS, ANSYS finally provided a simple 2-D

     solution which did not simulate behavior under a 3-D load input, but

     simulated the vertical drop of a box onto a fixed surface subject

     only to a gravity field.  The problem focused on the two gap

     elements that simulated the contact.  The response to be expected

     was that of a "pogo stick" with the rigid box bouncing in the

     vertical plane with the uni-dimensional motion preserved.  ANSYS

     demonstrated that this "perfect" or "idealized" non-linear

     simulation problem did lead to multiple solutions on two different

     platforms although it did not address whether a multiplicity of

     platforms would give only the same two differing solutions in the

     more complex fuel rack problem.  The conclusion suggested by the

     ANSYS simulation, prompted by the draft NRC notification, was that

     the inference that some of the gaps in the Holtec problem had the

     same initial value and that this created a "perfect" or "idealized"

     configuration which would be prone to multiple solutions.  ANSYS

     demonstrated their thesis by slightly perturbing the initial gap on

     one of the two elements in their simple model at time 0.0, and

     demonstrated that this eliminated the "perfect symmetry" and led to

     a single solution on both platforms.  I concurred with their

     conclusions that a perfect" or "idealized" solution could certainly

     lead to multiple solutions.  However, I did not concede that the

     fuel rack problems wherein the non-unique solutions were produced by

     ANSYS had the perfect symmetry of the 2-D problem.  Neither did I

     offer that the draft notice should be withdrawn.  It was my

     understanding that serious discussions were to be undertaken within

     NRC as to the necessity or advisability of a notice or suggesting

     additional clarifications by ANSYS as to the limitations of this

     code within their manuals.



5.   In subsequent conversations by phone with Dr. Bagchi on June 19,

     1997, I received the impression, perhaps erroneously, that the final

     resolution by the NRC not to issue the notice or to request some

     additional discussion by ANSYS was based on an internal NRC

     conclusion that ANSYS had sufficient general warning in their

     documentation about the potential pitfalls of non-linear

     simulations, and that Holtec had most likely erred by failing to

     recognize that they had a number of gap elements with the same

     initial gap which would potentially simulate a "non-realizable or

     perfect" problem.  Upon serious reflection subsequent to the June

     18, 1997 meeting, I felt that this issue needed to be addressed and

     upon further consideration, I do not believe that the ANSYS response

     really closes the issue.



6.   ANSYS makes much of the fact that the Holtec spent fuel rack

     solution using ANSYS involved a number of non-linear elements, a

     transient solution, no damping at the gap elements, and Holtec's

     choice of a "perfect" problem.  In reality, while the first two

     assertions are correct, the latter two are not.  The appropriate

     structural damping, mandated by the DOE specification, was present

     in the solution, and I had previously noted this fact to ANSYS

     during our initial queries on the problem.  Further, the fuel rack

     problem was not "perfect".  While it is true that some of the

     various gap elements in the model had initial gaps that were either

     0.0 or were chosen to reflect the reality of the



     simulation (i.e. a spent fuel rack initially at rest on a supposedly

     flat slab), the fact of the matter is that the model was real and

     truly represented an achievable initial state.  That is, there was

     no artificial construct in an attempt to simplify the solution.  Of

     most importance is the fact that the loading in the Holtec model was

     not symmetric and by itself served to provide the initial

     perturbation needed to drive the simulation to a unique solution.

     The input seismic loading had three components; therefore, no matter

     what "perfection" existed at time=0.0, at every time greater than

     zero, the gap elements experience different forces at all subsequent

     time instants.  This is markedly different from the ANSYS test

     solution where the problem should always respond in a completely

     symmetric manner as a single degree of freedom model throughout the

     time of the event.



On the basis of the above, my feeling is that the response given by ANSYS

which tends to absolve the code at the expense of the user,

(sophisticated or unsophisticated) is not so clear cut.  Either ANSYS

should stable the model request for a spent fuel rack dynamic analysis

simulation is too complex for implementation on its program, or take

steps to ferret out the still hidden interaction between gap elements and

the solution algorithm which makes their solution suspect.  I believe

that there considerations should be passed on to the USNRC.



*** END OF DOCUMENT ***