ML20149E933
| ML20149E933 | |
| Person / Time | |
|---|---|
| Issue date: | 06/06/1994 |
| From: | Buslik A NRC OFFICE OF NUCLEAR REGULATORY RESEARCH (RES) |
| To: | Strosnider J Office of Nuclear Reactor Regulation |
| References | |
| NUDOCS 9408080211 | |
| Download: ML20149E933 (5) | |
Text
'
Jur- 1994
@?T N01s. . . : Jack Strosnider, NRR/EMCB M#P 7M NW FROM:
' Arthur Busiik, RES/PRAB aw
SUBJECT:
CRITERIA FOR STEAM GENERATOR 17, 1994 TUBES, d subject document, Mr. Thadant stated that he w e who had expressed technical concerns with the voltage-based appro additional opportunity to provide coments on the draft generic letter also stated that he would appreciate receiving coordinated . He comen Imay have given my comments be in the coordinated comments from RES. to Mike Mayfield, but only anssummary o
comments, and their technical basis, directly to you.I am therefore transmitting my ,
The be used.GL states that only the log-logistic probability of leakage u curv curves should I on be the used.other hand believe that all'six probability of' leakage There' is nopoint from a theoretical or goodness-of-fit reasont of view.to choose one' curve over ano of view, since, as far as anyone knows, all of the curves could unde the true leakage, or all of the curves could overestimate the would that, one seem couldprudent use, say to theuse the most conservative offailing logistic the curves;true to do leaka which, I believe, tends to fall closer (to the middle of the calculated nge.
In addressing stated that: my concern (concern 2) in Enclosure,4itofisthe package log-logistic POL function (current industr"The genesic letter suitably conservative calculated leakage."y practice)ge as this provides a using onl There is no'possible conservativet way the logtlogistte POL function can o be beiconside the six curves investiIt generally gives the lowest or the next to lowest leakag curve over the others. gated, and there is simply no basis for choosing one The'"currenirindustry practice" observation ir meaninglessw-the plugging criteria. " practice" started during the considerations of interim There is no basis for its use There is no historical evidence that it is better than the other curves..
The addressing of my concern (concern 2) states that the leakage ca is very conservative when compared against the orich:al industry prop The relevant question is whether it is realistic, or conservative reality. .
e to relativ It could be the and stilleunderestimate conservative leakage. relative te the original industryproposal It is true that the 95% calculational bounding approach as given by the equation for h at the end of section 3.5 1
Coas
]ps 9400000211 940606 t PDR REVQP NRCCRCR CORRESPONDENCE PDR
of enclosure 3 of the GL package adds an element of conservatism. However, l the statement made on p. 5 of enclosure 3 of the GL package, in section 3.2,
{ that the calculated leakage is expected to exceed the actual leakage in 95 out of 100 cases is false. Important caveats on p. 3-21 of the NUREG-1477 draft for comment have been neglected. In particular, on p. 3-21 of NUREG-1477 it is stated: " h may be an underestimate if the Pf are underestimated."
l 1
That is precisely the situation I am concerned about.
1 Moreover, the log-logistic POL curve that is being used is the maximum likelihood estimate--not only is there no consideration of model uncertainty (which family of binary regression curves to use, the log-logistic or the Cauchy or another family of curves), but there is no consideration of paramater uncertainty for the parameters of the binary regression POL curve.
The proper way of" doing the calculation is to propagate all the uncertainties through. There is no other way of knowing how conservative, or how non- ,
conservative one's estimate of the leakage iss Incidentally, the parameter uncertainty is approximated by assuming the parameter estimates are normally distributed; this is true asymptotically, as the number of experimental points used to obtain the fit increases, but it is far from clear that this asymptotic limit is valid. It would be possible with some work to perform Monte Carlo calculations, along the lines of calculations indicated in the Numerical Recipes series (see, e.g., Press et al., Numerical Recipes in C, p. 548, 1st edition, 1988.) to verify the accuracy of the asymptotic normality approximation. I suspect that the parameter uncertainties are underestimated. (Moreover, one must always remember that the parameter uncertainty calculations assume the validity of the particular model used--whether Cauchy, log-logistic, or some other model.)
In enclosure 4 of the GL package, in the discussion of Concern 2, it is mentioned that hydrostatic leakage tests performed by EDF indicate leakage significantly approach. less than currently calculated with the voltage repair limit But as noted on p. 5 of Enclosure 3 there are limitations to the French hydrostatic test data. They are performed under static conditions at room temperature.
It is unlikely that NRC' consensus ort the applicability of the French hydrostatic tests can be obtained. Thus one cannot depend on the French hydrostatic tests ast support for:conservatise irr the leakage-calculationss An additional alleged aspect of conservatism in the leakage calculation mentioned in the GL enclosure 4 with regards to Concern 2 is the use of a probability of detection to take into account the non-detected indications.
This may result in a higher estimated leakage than the original industry calculations, but it does not mean that the results will overestimate the true leakage. One should use a realistic probability of detection curve, as best as one can.
2
l Thus there is' no basis for the assertion that the estimated leakage is conservative in the sense of exceeding the true leakage. The' estimated leakage could easily be less than the true leakages 1
The statement made in section 3.5.2 of enclosure 3 of the GL package that any non-conservatism associated with the use of a log-logistic model as compared to other functional forms is small compared to the conservatisms 1nherent in the existing methodology cannot be supported. Although it may be true that the range of variation of leakage was only a factor of 2 to 4 when one used the different probability-of-leakage models in the cases examined, and when
) be the case in the future. compared only the maximum likelihood estimates, it is not c j
plugged may differ considerably for the different curves.Moreover, the number On p.16 of enclosure 3, in section 3.5.2, mention is made that at I volt the i POLthe and for3/4 theinch log-logistic diametercurve tubing.is less than 1% at 1 volt for both the 7/8 inch -
J when the other binary regr ession models are used.No mention is made of what the res From Table 7 of the report supplied as part of technical concern 2 in enclosure 4 of the GL package it can be seen that, for 3/4-inch tubing, although the log-logistic curve gives a maximum likelihood estimate of .5% for the probability of leakage at 1 Volt, for the curve Cauchy curve the corresponding value is 3.2% and for the logistic 1.8%.
uncertainty. In addition, the results given take no account of tne parameter basis assumptions for calculating offsite dose releases res conservative calculation of offsite releases.
in the iodine spiking model and in the iodine activity. Mention is made of conservatism In point of fact, the licensing model'for iodine. spiking.can plausibly seriously underestimate the releases, for a steam line break accident. The reasonSoc.
Nucl. is that, L8, according to a plausible model by Lutz and Chubir (Trans. Am.
- p. 6491978)), the iodine spike release from the clad-pellet gap on reactorto the primary coo (lant occurs when the gas in the clad-pell depressurization.
The data on iodine spiking is taken for normal shutdowns and scrams, where the primary coolant depressurization occurs over ,
some hours, and under these circumstances the release of iodine from the clad-pellet gap to the primary coolant occurs over a period of a few hours.
However, in a main steam line break accident the depressurization occurs over a few minute time span, because of cooling down and shrinkage of the primary coolant.
The model of Lutz and Chubb is as follows. Iodine is present during normal operation in the fann of low volatile iodides (probably cesium iodide) on the surface of the fuel pellets and in fuel crack zones. When the reactor shuts down, water vapor in the fuel-pellet gap condenses, and the iodides on the surface of the fuel pellets v ssolves in the liquid water. During coolant depressurization, gas in th.< ,,ellet-clad gap expands, forcing the iodine-laden water out into the coolant. According to investigations of spent fuel rods (see Neeb and Schuster, Trans. Am. Nucl. Soc. 28, p.650(19'!)) there is about 130 Ci of I-131 iodine deposits accessible for condensed water in each fuel I 3
i
.. rod. With a model like this <
the iodine spike occurs over,a time scale depending on the rate ofthe!
depressurization, and for a rapid depressurization such as occurs in a main <
steam line-break awill depressurization, occur few over a time scale corresponding to the minutes. )
I Using steam line such a model, we can estimate the dose to the thyroid during a main break. I defected rod, mixes' instantaneously with the' primary coolant.W ass I two hour We consider the by dose to the thyroid at the site boundary. The dose to the thyroid is given D=De g'B(X/0) T l
where 0,,
is the dose conversion factor from Ci to rem, 8 is the brea release to the environment, averaged over the time T, and X/Q is the dispersion factor, per unit source. giving the atmospheric concentration at the site boundary The value of Q' is given by O'=p C2 F where p the primary coolant water density, C, is the iodine coolant concentration, and F is the primary to secondary volumetric flow rate. All' of the iodine passing from the primary to the secondary is assumed toipassi directly from theto theof'the time environment main steam6-the line,affected break. steam generator is assumed dried out rate pF is assumed constant at its initial value.In evaluating Q', the mass flow F is measured in gallons per minute, p in g/cc, and CIn the above in Ci equation,
)er gram, a if conversion factor of 3780 (the number of cc in a gallo,n) is required on the right hand side; the result will then be in Ci/ min.
The concentration of iodine in the coolant based on the above model of instantaneous primary coolantmixing is givenofby the iodine released from the pellet-clad gap to the C,=Oc /H,=0,a f ef ,a,/H, where Q, is the total amount of iodine in the coolant, Q.=130 Ct.is the iodine released from each defected rod, f., is the clad defect level, giving the infraction rods the core, of androds in mass M is the the core which coo of the primary areant.
defected, ComputationsN,l,, is the num were leak rate done for30 F of a clad gpin,defect level of.005 (.5Wh and a primary to secondary A value of 1500 regwas obtained for the two hour dose 4
I. i to the thyroid, far in excess of the 30 rem required by the standard review plan. The primary coolant concentration of iodine obtained, C,, was colculated ast165pC1/gW. The input data, in addition to assumed clad defect level of .005 and the leak rate of 30 gpm was:
M,-2E5 kg (corresponds to a volume of 10000 cu. ft at a density of .7189/cc)
N,,,,-51000 .
p=.718 g/cc B .000347 m'/s D ,-1.48E6 rem /Ci T-120 minutes X/Q .0018 s/m'(2 hour2.314815e-5 days <br />5.555556e-4 hours <br />3.306878e-6 weeks <br />7.61e-7 months <br /> dose)
The results are proportional to the clad defect level and the primary to secondary mass flow rate.
It seems clear that it is important to obtain whatever data is available on iodine spiking from cases where rapid primary system depressurization have occurred, if any such cases are available. Especially important would be the data on the timing of the release from the pellet-clad gap to the coolant, under rapid system depressurization such as would occur on a main steam line break. During normal shutdowns and scrams it is conservative to assume that the release takes place over a two hour period, but the above model, if valid, would predict that the release from pellet-clad gap to coolant would occur over a much smaller time period, for cases of rapid primary system )
depressurization.
There is an implication in the GL package that a reduced primary coolant iodine activity will result in a proportionally reduced release rate of iodine i
to the coolant under iodine spiking conditions. This would be valid if the I model used in the standard review plan section 15.1.5, Appendix A, were valid.
However, inspection of the data on iodine spikes given in Adams and Atwood, Nuclear Technology 9_4, p. 361 (1990) does not, at least at f.irst glance,-
support this. One possible reason for this lack of non-prop 6rtionality would be that some of the primary coolant iodine during normal operation comes from i a source other than that associated with clad defects, such as the tri.:ap !
uranium on the surface of the cladding which is mentioned in the introduction to the Adams and Atwood paper. Of course, if the primary coolant iodine specific activity is reduced by increasing the purification system flow rate, this would have no effect on the iodine spiking release rate, but would result in a lower primary coolant iodine activity during normal steady state power conditions. The degree to which a reduction in the technical specification value of the primary coolant iodine activity affects the iodine spike release deserves a closer look.
I l
\
5
_ _ __.