HNF-3233

Analysis of SX Farm Leak Histories --Historical Leak Model (HLM) August 1998

III. Methodology:

The SX Farm tanks were used to store high-level waste from the Redox Process (S Plant) with very large peak heat loads, upwards of 1-2 million Btu’s/hr. These heat loads were largely a result of the short-lived radionuclides that were present in the fuel that was being processed. After approximately six years these early heat sources decayed leaving the 30-year isotopes Cs-137 and Sr-90, which represent only 0.5% of the heat load at 150 days cooling time. During the periods of extremely high heat load, tanks were cooled by reflux and large amounts of water boiled from the tank were returned as condensate. At peak heat loads, upwards of 200 kgal of waste inventory was evaporated per month of operation for some tanks and at peak times the rate amounted to some 450 kgal of condensate collectively for the entire tank farm. During some of these peak boil-off time periods, leaks on the order of 0.5 to 2.0 kgal per month appeared in many tanks and it would have been very difficult to discern these leaks as unaccounted material in the presence of such large volume turnovers from self boiling.

The Historical Leak Model attempts to reconcile reported tank volumes with those calculated based on expected volume reduction associated with historical heat loads for each tank’s waste. The HLM calculates volume changes based on the evaporative cooling rate of a tank and reconciles those level changes with reported volumes and transactions for each tank during the same time period. Using this reconciliation, measured tank volumes that are greater than those expected because of evaporative loss are accounted. Measured tank volume losses less than those predicted by waste evaporation are assigned as unaccounted volume losses. These unaccounted volume losses may be due to normal volume loss, inaccuracies of the HLM, or waste inventory that has been lost to the soil through a breach in the tank liner.

For example, in January 1961, the evaporation rate for SX-108 was 2.0 kgal/mo and the unaccounted volume loss was 2.0 kgal/mo. This assumes a total tank heat of 38 kW, a heat loss to ground and air of 20 kW leaving an 18 kW evaporative loss. In principle, the leak rate could be as much as a factor of two lower, on the order of 1 kgal/mo, given a higher evaporation rate and a corresponding lower heat loss to ground. If the heat loss to ground were as low as 10 kW, the evaporation rate would be 50% greater (3.1 kgal/mo) lowering the leak rate to 0.9 kgal/mo. A critical factor is the tank waste temperature during this period. If tank waste temperature data show a significant decrease during this period, this will also lower the leak estimate. A decreased tank temperature suggests that evaporation is cooling that tank waste beyond what is necessary to remove the tank radionuclide heat load.

Tanks that were cooled by evaporation are assumed to have been on a reflux condenser system. In a total reflux, condensate derived from tank vapor is returned to each tank from a cooler. Although the details of this process are unclear at this time, generally the condensate was extracted from a number of tanks with a common ventilation system though separate isolated coolers were also available for individual tanks as well. In fact, specific mention is made for placing SX-107 and SX-108 on isolated coolers during periods of uncertainty about unaccounted volume losses for these two tanks. When tanks were on common ventilation, though, it was not possible to determine how much condensate actually came from each individual tank in the entire group.

In addition, there was inevitably some loss of water vapor out of the system during the process and this loss needed to be made up by periodic water additions back to the tanks. It is only when this added makeup volume exceeded some specified tolerance that any leak would be discernible from an accounted materials basis. Thus, a leak that was small on the order of the system throughput would be completely masked by the operation.

If a tank was allowed to self-concentrate, condensate would be directed to SX-106 instead of being returned to the tank. Often the condensate was partly refluxed and partly extracted, resulting in net concentration of a tank’s waste inventory to some target volume. This condensate in SX-106 presumably was used for makeup additions. During this process, there was undoubtedly some loss of water vapor to the air and that loss needed to be made up by periodic additions of water back into each tank. This amount of makeup would likely scale with the total reflux rate of each tank. For example, a 2% loss would mean that SX-108 would need a 2.6 kgal/mo makeup in October 1963, since the reflux was evaporating 130 kgal/mo at that time. The unaccounted volume loss during that period was on the order of 1.5 kgal/mo and therefore the unaccounted volume was more than one-half of the entire makeup volume. The amount of unaccounted volume relative to the makeup volume is very important since it is the makeup volume that masks small tank leaks. Moreover, since the entire farm’s evaporation rate was on the order of 200 kgal/mo, such a leak would have been effectively masked by the normal water makeup for the process.

The HLM assigns leak rates on the basis of unaccounted volume losses during periods of low reflux. Extrapolation of those leak rates into the future produces bounding estimates for total inventory losses for the duration of each leak. Therefore, it is important to consider the causes for tank leaks in order to better determine their potential start dates.

Leaks often appear on the tails of tank heat load curves. It is not clear at this time if this is because there was some physical mechanism that took place with tanks during these periods or if that means that the high reflux rates for these tanks prevented any effective determination of unaccounted volume losses. The evaporation rate for a tank’s waste is an increasingly sensitive function of the tank temperature and ventilation rate during these times. The HLM assumes that the tank heat loss to ground and air are constant during these times at 10 kW each for a total loss of 20 kW.

Bounding estimates for leaks from waste tanks depend critically on the timing of a leak as well as on its size. If a leak begins during times of high thermal stress, which is a reasonable assumption, it will be masked by the correspondingly large reflux rates and most likely will not be evident until the reflux rate of a tank is reduced considerably . Only then will the leak rate become a significant fraction of the normal makeup volume. Moreover, tanks connected to a common ventilation system will be subject to the reflux rate of the entire system, not just that for individual tanks. This interlinking of the tanks would have seriously complicated any determination of volume unaccounted for an individual tank.

Reasons for tank failure:

Stress-corrosion cracking failure has been most often blamed for the majority of tank leaks. It begins at welds that have residual stress and propagates as cracks at right angles to the welds. Thermal stress presumably contributes to increasing these corrosion rates and many single-shell tanks were subjected to temperatures on the order of 140-160°C (280-320°F) for sometimes ten or more years. Newer double-shell tanks were stress relieved prior to use and none have shown any sign of leaking. On the other hand, the DST’s have not been subjected to the very high thermal stresses and severe steam bumps that were common for SST’s in S, SX, A, and AX Farms.

Undoubtedly stress-corrosion cracking has played a major role in many tank leaks, but high thermal stresses leading to sudden structural failure cannot be neglected as a potential explanation for tank leaks for SST’s with large thermal loads. This mode of failure is certain for at least three tanks at Hanford: SX-108, SX-113, and A-105. Severe bottom bulges were observed for each of these tanks and directly associated with their leaking waste into the soil column. For both SX-108 and SX-113, the bottoms relaxed back to normal some years after they bulged. It is very likely therefore that other high heat tanks also suffered bottom bulging that also relaxed back and therefore were never observed. Some eleven other leaker tanks that were thermally stressed are: A-103, A-104, S-104, SX-104, SX-107, SX-109, SX-110, SX-111, SX-112, SX-114, and SX-115. In fact, bent temperature probes, which are highly suggestive of past deformation of a tank bottom, are present in two of these tanks: A-104 and SX-109.

It is clear then that sudden structural failure at least played a role in tank failures at Hanford. The question becomes how can we use that information to help define the tank leak amount and timing.

Another factor to consider is the total volume of soil associated with the leak. Drywells are positioned 10’ from the tank edge and for tank SX-108, 08-11 and 08-02 both show activity centered on 55’ depth. Assuming that the entire soil quadrant is saturated to 10’ from the outer edge of the tank, this would require 4.4 kgal waste per foot of saturation depth with a 20% porosity of the soil. Thus, a 203 kgal leak should have a saturation depth of 46’ .

Leak Model:

Each leak is described by three parameters, a leak elevation (in equivalent kgal tank volume), a leak rate (in kgal leaked per month per kgal head above the leak elevation), and a leak period. Thus, a leak is assumed to derive from a particular elevation within a tank and therefore a tank only leaks when the waste level is greater than the leak elevation. The leak rate then increases linearly as the hydraulic head above the leak elevation increases. Finally, each leak has a starting time and a duration. Note that since tank waste elevation and hydraulic head drive the leak rate, the resultant leak rate may vary substantially over its duration depending on the changes in the waste level within the tank.

A. leakVol(mo_i) = leakSize * (calcVol - leakElev)

B. leakAcc = S_i leakVol(mo_i)

In addition, there are numerous reports of leaks that "self-sealed" after they had begun and then perhaps later restarted. As a result, another potential variable is the leak "duty-factor" or that fraction of time that a leak actually occurred over the entire leak duration. However, there is simply not enough information for all these tanks to justify yet another leak parameter and so leaks for SX-108, SX-109, and SX-111 will be assumed to be continuous for their duration with a rate dependent only on tank elevation and hydraulic head.

For tank SX-112, there is a period of unaccounted loss in 1958-9 that suggests a loss of around 1.6 kgal/mo for 12 months. This is followed by a second cooling period 1964-66 that did not show unaccounted losses and then a final cooling period in 1969 where unaccounted losses occured once again. As a result, we have included two separate leaks for this particular tank with the assumption that the leak effectively sealed in between these events.

The difference in slope between the two lines (calc. vol. and vol w/o leak) for tank SX-108 (Fig. A-1) from 1960 to 1962 represents the leak rate for this tank. In other words, the tank heat load at this time was not sufficient to explain the reported waste volume losses. Likewise, these differences in slope define leaks for SX-109 (1967-9) and SX-111 (1973-4). For tank SX-112, there is much more limited data for its second leak and that leak is defined by only two months in 1969.

Soluble radionuclides are assumed to all follow Cs-137 and this is the only isotope calcuted by the HLM. All other radionuclides are calculated by scaling results from the HDW model estimates (see below).

Reflux model:

During the reflux process, each tank’s dome vapor was routed by an underground header system to a single or centralized condenser. Water was extracted from the tank dome vapor down to the dew point of the condenser temperature, which we presume was on the order of 65°F (18°C). However, this condenser temperature does not affect the HLM. The chilled air was then either vented to the atmosphere or returned to each tank. The condensate from this process was then returned to each tank as needed to maintain some target level for that tank along with some makeup volume. However, since it was often the case that many tanks were connected to a common ventilation system, the exact amount that needed to be returned to each tank could only be determined by the losses that were observed for that tank.

This meant that if some other loss pathway existed for the tank waste, for example a leak, this volume would also need to be made up. By measuring the makeup volume needed over and above the condensate extracted, one could then surmise if there were losses over and above those expected from makeup volume and take appropriate action to isolate which tank was leaking. Unfortunately, the normal steam vapor losses for the system must have scaled as something like the total reflux volume and these losses needed to be made up as well during the process. As a result, the normal makeup volume for the total system undoubtedly masked many individual tank leaks to the soil column.

Since the level record of each tank is mostly expressed quarterly or semiannually, there are substantial gaps in the volume record and those gaps would need to be filled to actually simulate each tank adequately. In particular, we do not believe that the tank waste volumes decreased to the extent represented by the HLM over a period of a quarter. The HLM does show the intermediate volumes decreasing substantially during many quarters and we presume that the actual decrease was limited to some more reasonable amount. For example, during 1956-60 we have monthly reports for these tank levels as well as monthly reports of concentration and water additions.

The amount of waste evaporated was calculated as

C. evapRate (in/mo_i) = tankHeat(kW) / waterHeatOfVap(kW/(in/mo))

where

D. waterHeatOfVap = 8.96 kW/(in/mo)
         (from heatVap = 2259 J/g x 3.785e6 g/kgal x 2.75 kgal/in / 2.625e6 s/mo)

is simply that of pure water at 100 C. The spreadsheet calculated total volume in kgal for each tank was

E. calcVol(mo_i) = calcVol(mo_i-1)
         + tankTrans(mo_i) - 2.75*evapRate(in/mo_i) + accWater(mo_i) + unaccWater(mo_i)

where there are 2.75 kgal/in and unaccounted water addition was based on

F. unaccWater(mo_i) = measVol(mo_i) -
         - (measVol(mo_i-1) + tankTrans(mo_i-1) + accWater(mo_i-1) + unaccWater(mo_i-1)
         - evapVol(mo_i-1) - leakVol(mo_i-1)).

Likewise, the tank volume in the absence of a leak was simply

G. noLeakVol(mo_i) = calcVol(mo_i) + leakVol(mo_i).

Note that calcVol and measVol coincide by adjustment of unaccWater in an iterative spreadsheet calculation. Thus, any changes in leak or other parameters cause the spreadsheet to recalculate to a new set of unaccWater additions for that tank. This calculation takes several minutes and is iterative since changes in unaccWater for a given month affect all subsequent months.

IV. Results:

Tank leak estimates are primarily derived by examination of unaccounted volume losses, but we use the lateral and drywell contamination reports to corroborate leak start times as well. We attribute all unaccounted volume gains to water additions that replace each tank’s evaporative loss.

Tanks in SX Farm were cooled primarily by reflux during this time. The reflux involved extraction of steam from tank dome space followed by contact with a condenser to condense and remove water. The condensate from this process was returned to the tank and the air was either vented to the atmosphere or returned to the tank. Since tanks with very high reflux rates lost tank inventory at very high rates (up to 200 kgal/mo), there were correspondingly large uncertainties in unaccounted volume losses associated with this reflux cooling of each tank. Most of the inventory loss during reflux was immediately returned to the tank as water, but some undoubtedly also escaped as steam during the operation. This condensate loss had to be made up by periodic water additions and the amount of this makeup could effectively mask relatively small losses due to leaks.

Many tanks were connected to a common ventilation system in SX Farm which would have further complicated the already difficult determination of material balance. A tank that developed a leak contributed an unaccounted volume loss to an entire group of tanks that were on a common ventilation system. Therefore, a leak could be masked not only by its own high evaporation rate but also by other tanks within a group with high boil-off rate tanks.

We know that 242-T and 242-S in-farm evaporators used a 5% materials accountability criterion for their operation [see evaporator references]. Because the in-tank evaporation for SX Farm was much more complex than the in-farm evaporators, this suggests that SX Farm condenser operation probably used a materials balance on the order of 5-10%.

Table 2. Comparisons of HLM with other leaks for SX-108, SX-109, SX-111, and SX-112.

Hanlon is Hanlon, B.M. "Waste Tank Summary Report of Month Ending May 31, 1996," WHC-EP-0182-99, August 1996.

Anderson is Anderson, J.D. "A History of the 200 Area Tank Farms," WHC-MR-0132, June 1980.

Table 3. HLM leaks for SX-108, SX-109, SX-111, and SX-112.

*decayed to 1/1/94

Notes on specific tanks:

SX-108

Tank SX-108 showed its first HLM unaccounted volume loss in 1959 (see App. A, column "unacc.vol.", third quarter 1959). The HLM leak assignment begins at this time as shown in Fig. A1 and column "kgal leak". A bottom bulge may have appeared as early as 1959 but a tank bottom bulge was reported in September 1967 and subsequently relaxed back to a flat position [Brevick, et al.]. There are bent temperature probes in the northwest quadrant of this tank yet today, which is about where the bulge was reported previously. The drywell with the greatest activity is 08-11, which is adjacent to the northwest tank quadrant, once again the same quadrant that bulged. In addition, all three lateral wells underneath this tank show large amounts of activity.

Although there were reports of lateral well activity in SX-108 as early as December 1962, the tank was emptied and refilled with another round of fresh Redox waste from mid 1962 through 1964. This tank was placed in reflux during this time with a peak evaporation rate by the HLM of 132 kgal/month (48 in/mo of waste inventory) in late 1963. By mid 1966, the reflux rate of the tank had slowed significantly and the tank once again showed a steady loss in volume over what was expected based on its HLM heat load.

Concern with increased lateral well activity in November 1965 dictated that the tank be placed on an isolated cooler and over the next several months, the tank’s level was reported to have stabilized and the leak therefore assumed to have "self-sealed." The level data within the HLM, though, shows no such level stability at that time. In fact, the HLM SX-108 inventory decreased 13 kgal from the December 1965 of 653 kgal to the June 1966 level of 640 kgal. The reflux rate predicted by the HLM in November 1965 was 20 kgal/mo while the HLM leak rate at that time was 1.6 kgal/mo. It is not clear that such a leak rate would have yet been evident for this tank given this rate of boil-off and so a continuing leak is possible during this time despite the reports of the tank’s leak having "self-sealed."

Maximal and minimal leak volumes are derived by assuming a continuous leak with the leak parameters shown in Table 3 (maximum leak rate was ~3 kgal/mo and varied according to waste level). Total volume loss up through 1962 was 203 kgal if the leak is assumed to have continued through 1967. These leak amounts depend critically on assumptions of heat losses to ground and air as well as on the continuous nature of the leak. Therefore, each leak estimate may be ~50% lower due to inaccuracies in the HLM for low heat load tanks. No account is taken in this analysis for potential "self-sealing" or non-continuous leaks. Unaccounted volume losses after 1967 may have been another leak or simply an extension of the old one, but since the waste within the tank was solidifying at that time, these level changes become increasingly difficult to interpret. Considering the fact that some inventory in the tank seems to spontaneously appear in 1969, we attribute this later unaccounted volume loss to a measurement anomaly and accordingly ignore it.

A leak volume of 203 kgal represents 27,100 cu.ft. of waste which would wet approximately 135,000 cu.ft. of soil at 20% porosity. This volume of soil is a sphere approximately 64’ in diameter and therefore would imply a lateral penetration into the soil column underneath the tank on the order of 32’.

Figure A2 compares the rates for unaccounted volume gains and losses with those for leak and evaporation for the period 1959 to 1962. We derived the leak rate for SX-108 by adjusting it to accommodate as much of the unaccounted volume losses as reasonable. This produces the leak rate shown in Fig. A2 as well as a set of unaccounted volumes that now represent no change in volume or volume gains (solid diamonds). Fourth quarter 1960 still shows an unaccounted volume loss, but we felt that this quarter’s large unaccounted volume loss was beyond that that the HLM could reasonably accommodate. This volume loss may represent a variable leak rate or it may represent the uncertainty of data within the HLM.

SX-109

Tank SX-109 shows its first unaccounted volume loss in 1960 whereas Hanlon reports 1965, Anderson reports December 1967, and other sources report January 1965. Significant long term unaccounted volume losses continued for many years for this tank. The HLM shows this time period to have had insufficient heat load to explain all long-term volume loss by evaporation. These unaccounted volume losses amount to 111 kgal over these many years. This amount may once again be as much as 50% too large because of HLM inaccuracies for low heat loads. Thus, a lower bounds would be 56 kgal leak for this tank. This is a tank that shows clear evidence of a very slow leak over a long period of time, which is consistent with the fact that the drywells associated with this tank were showing readings consistent with continued migration of radionuclides as late as 1976. There are bent temperature probes in the north edge and southeast quadrant of this tank but no bottom bulge was ever reported for SX-109. The drywell 09-04 that is adjacent to the southeast quadrant does show activity associated with a leak.

Figure A4 compares various volume rates for SX-109. The leak rate for this tank was adjusted to accommodate all unaccounted volume losses. Note that the leak rate reduces significantly in late 1960 as a result of the fact that leak rates scale with tank volume and tank volume changed at this time (see Fig. A3).

SX-111

Tank SX-111 shows an HLM unaccounted volume loss in third quarter 1972 of about 2.4 kgal/mo. Extending this leak until the tank was pumped in second quarter 1974 amounted to a total of 55 kgal for this leak duration. Since this unaccounted volume loss appeared at a time of very low heat load, it is subject to increased uncertainty because of the HLM deficiency with low heat loads in tanks. During the period of the leak, the evaporation rate decreased only slightly from 5.5 to 5.0 kgal/mo while that of the reflux coupled system dropped from 15 to 12 kgal/mo. Thus, this leak estimate may be as much as a factor of four too large, giving a 15 kgal lower limit for this leak.

At any event, lateral and ground contamination in 1974 resulted in this tank being assigned as a leaker. Tank SX-111 has three other periods in the tails of cooling curves where a leak should show up; one in 1958-9, the second in 1963-4, and a third in 1969-71. Neither of these periods show indication of unaccounted volume losses by the HLM, which suggests that the HLM is reasonably well calibrated. Figure A6 shows the comparison of volume rates for SX-111 for the period 1972 through 1974. Since there is only one quarter showing an unaccounted volume loss there is a great deal of uncertainty is associated with this leak assignment.

SX-112

Tank SX-112 shows its first HLM unaccounted volume loss in October 1958 with several additional months unaccounted losses in 1959 as well. At this time, boil-off was down to 1-2 kgal/mo while the unaccounted losses ranged up to 4 kgal/mo, suggesting a leak of around 1.6 kga;/mo over 12 months for a total leak volume of 19 kgal. These adjustments in unaccounted volume are shown in Figs. A8 and A9.

However, this tank was only noted as a leaker in 1969 as shown in Figs. A9, which is consistent with the assignment of this tank as a leaker by other sources. Figure A9 shows the volume rates for SX-112 where we have only a single quarter of unaccounted volume loss, first quarter 1969. This tank was pumped to heel in the second quarter and therefore we have a very limited duration for this leak with a loss of 25 kgal over 2 months or about 13 kgal/mo. At that time, SX-112 was evaporating at the rate of 16 kgal/mo while the system rate reflux was 50 kgal/mo. Since this leak was detected very soon after the unaccounted volume loss, we assume that the drywell contamination was evident at this time as well. Currently, we estimate that the total amount leaks was 44 kgal from both leak periods but that this may be as much as 50% too high or low as a result of the possibility of variation of either leak period.

SX-105

Tank SX-105 has not been assigned as a leaking tank and it is the only SX Farm waste tank outside of the leaking set, SX-107 through SX-115, that has lateral as well as vertical drywells. Therefore, it provides an ideal non-leaking tank with which to test the limits of the HLM. Application of the HLM to this tank is shown in Figs. A10 and A11. The 1957-63 cooling period shows only three calendar quarters of very slight unaccounted volume losses of 0.5 kgal/mo or less as shown in Fig. A11. This size of unaccounted loss represents the limit of leak that the HLM can discern. This leak rate corresponds to a leak size of 7.6e-4 kgal/mo per kgal head, or about a factor of three less that for SX-109 and a factor of six less than the leak sizes for SX-108, SX-111, and SX-112. We assume that the HLM simply is not sensitive enough to discern leaks of this size.

Continue to next section