ML13323B198
ML13323B198 | |
Person / Time | |
---|---|
Site: | South Texas ![]() |
Issue date: | 08/19/2012 |
From: | Howe K Southern Nuclear Operating Co |
To: | Office of Nuclear Reactor Regulation |
Shared Package | |
ML13323A673 | List: |
References | |
GSI-191, NOC-AE13003040, STI 33762096, TAC MF2400, TAC MF2401 CHLE-010, Rev 2 | |
Download: ML13323B198 (48) | |
Text
NOC-AE-1 3003040Attachment 5CHLE-010:
CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition PROJECT DOCUMENTATION COVER PAGEDocument No: CHLE-010 TRevision:
2 Page 1 of 32Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum AdditionProject:
Corrosion/Head Loss Experiment (CHLE) Program Date: 19 August 2012Client: South Texas Project Nuclear Operating CompanySummary/Purpose of Analysis or Calculation:
Corrosion/Head Loss Experiment (CHLE) tests are being performed to support the risk-informed resolution of GSI-191 at the South Texas Project Nuclear Operating Company (STP). This documentpresents the results of two multi-day tests performed in the CHLE test system (incorporating both the tankand the head loss assemblies) that evaluated methods for preparing the fiber beds for the 30-day tanktests. Two fiber preparation methods were tested: (1) fine chopping of fibers in a blender, and (2)separation of fibers using the NEI pressure-washing method.To test the response of each fiber preparation method to the presence of chemical
- products, aluminumnitrate was added very slowly in successive batches over several days. The aluminum nitrate additionbegan after about 7 days of circulation under fiber only conditions.
Role: Name: Signature:
Date:Prepared by: Kerry Howe < signed electronically
> 7/23/2012 UNM review: Janet Leavitt < signed electronically
> 8/17/2012 STP review: Ernie Kee < signed electronically
> 8/17/2012 Soteria review: Zahra Mohaghegh
< signed electronically
> 8/18/2012 Revision Date Description 1 7/23/2012 Draft document for internal review2 8/19/2012 Addressed internal review comments Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum AdditionTable of ContentsIntroduction
.....................................................................................................................................
3Sum m ary of Results ........................................................................................................................
4H ead Loss through Fiberglass Debris Beds ................................................................................
6Approach V elocity through Fiberglass D ebris Beds .................................................................
8Tem perature
..................................................................................................................................
10Tem perature Profile over Tim e .............................................................................................
10Tem perature V ariation in Tank .............................................................................................
10Tem perature D ifferential in Head Loss Colum ns ..................................................................
10Bed Form ation and M orphology
...............................................................................................
13W ater Chem istry ...........................................................................................................................
18p H ..............................................................................................................................................
1 8Calcium and Silica ....................................................................................................................
18Effect of A lum inum A ddition ...................................................................................................
21Turbidity
...................................................................................................................................
21A lum inum Concentration
......................................................................................................
21Total Suspended Solids ........................................................................................................
24D ebris Bed Head Loss ..........................................................................................................
25Particle Size and Zeta Potential
............................................................................................
28Conclusions
...................................................................................................................................
30References
.....................................................................................................................................
32Document No: CHLE-010, Rev 2 Page 2 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum AdditionIntroduction The purpose of this report is to describe the results from experiments conducted as part of theCorrosion Head Loss Experimental (CHLE) program.
The CHLE tests are being conducted atthe University of New Mexico to investigate long-term chemical effects on Emergency CoreCooling System (ECCS) strainer debris beds under prototypical conditions for the South TexasProject in support of the NRC Generic Safety Issue (GSI) 191 risk-informed resolution.
Twoobjectives within the CHLE program are to determine (1) whether or not chemical precipitates can form in the post loss-of-coolant accident (LOCA) environment, and (2) whether anyobserved products either nucleate directly on or accumulate within prototypical fiberglass debrisbeds.An important consideration is that the fiber debris beds that are used in the chemical effecttesting be suitable surrogates for debris that would be formed during a LOCA. Attributes thataffect the suitability of a particular debris bed design include the stability of the debris bed, thereproducibility of the results, and the ability of the debris bed to participate in chemicalinteractions under a variety of conditions.
Debris beds used in some previous GSI- 191 work arenot necessarily applicable to the current study for three reasons.
First, the approach velocities historically used in head loss testing were more than an order of magnitude higher than the STPstrainer design. Second, the historical observations were typically for short periods compared tothe CHLE investigations.
Third, the Nuclear Energy Institute (NEI) has recently developed adebris preparation method [I] that is believed to be prototypical of debris formed during aLOCA, and most previous head loss testing have used other debris preparation methods.
TheNuclear Regulatory Commission (NRC) reviewed the NEI plan and noted it is generically anacceptable way of producing debris, but declined to officially endorse it as the only way toproduce acceptable debris because of the dependence on human actions [2].Two types of fiber bed preparation methods have been evaluated for possible use within theCHLE program.
First, the most recent debris formulation advocated by NEI for strainer testinginvolves baking fiber blankets on one side at 300 'C for 6 to 8 hours9.259259e-5 days <br />0.00222 hours <br />1.322751e-5 weeks <br />3.044e-6 months <br />, followed by disaggregation with a commercial pressure-washer; this method is referred to as the NEI pressure-washing method in this report. Second, fiber blankets were subjected to the same backing procedure, butwere separated by fine chopping of fibers in a blender.
A previous report, CHLE-008:
DebrisBed Preparation and Formation Test Results [3] performed an initial investigation of thesedebris bed preparation methods with respect the attributes for suitability described above. Theexperiments described in that report found that the NEI pressure-washing method resulted inmore stable and reproducible debris beds than the blended bed method in relatively short (severalhour) head loss tests. However, the blended debris beds experienced greater head loss whenprecipitates prepared according to the WCAP protocol were introduced directly into the headloss assemblies or into the CHLE tank, leading to the perception that the blended fiber debrisbeds are more sensitive detectors for the presence of precipitates.
The current test results extend the knowledge about the suitability of these debris preparation
- methods, and provide additional information about the formation of precipitates in theprototypical chemical environment.
The tests had two key components.
First, the stability andreproducibility of the debris beds were investigated over a longer period (about 7 days) toDocument No: CHLE-010, Rev 2Page 3 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Additionevaluate whether the debris beds would be suitable for the long-term CHLE tests, which may beup to 30 days long. This portion of the study was conducted with no corrosion materials in thetank or corrosion products added to the solution.
Second, after 7 days of fiber-only operation, aluminum nitrate was slowly injected into the tank to simulate the slow release of aluminum thatwould occur as the result of corrosion.
This aluminum addition method is believed to be a closersurrogate for corrosion product introduction than direct addition of the WCAP formulation.
The key objective for selection of bed morphology in the CHLE program is that the debris typebe representative of nominal beds that are likely to accumulate within the spectrum of breaksizes; i.e., the bed should not be artificially constructed.
The tests described here explore theattributes of two alternative bed preparation methods; both of which might be considered realistic under different recirculation flow regimes, but neither which can alone fully inform theresolution of chemical induced head-loss affects.
Therefore, practical considerations regarding test stability for the purpose of studying 30-day chemical behavior is a dominant concern for theselection of a debris preparation protocol.
Head loss through debris beds involves a wide varietyof physical phenomena.
These phenomena include initial debris size, fiber separation, and fiberfracture, prototypical debris transport, and accumulation.
None of these factors affects thechemical behavior of fiberglass in the system, but all of these factors affect the degree of headloss that can be experienced at the strainer.
Issues like particulate to fiber ratio, maximum bedthickness, thin bed formation under quiescent flow conditions, etc. will be studied systematically in the vertical head-loss test series.The testing program was conducted from 28 June 2012 to 24 July 2012. Throughout the tests,the chemical system in the tank was prototypical of the post-LOCA chemical environment atSTP; the chemicals included boric acid, trisodium phosphate, lithium hydroxide, hydrochloric acid, and nitric acid. The support screen use was prototypical of the ECCS strainers at the STPplant. A temperature profile characteristic of a medium-break LOCA as predicted by MELCORand RELAP-5 was used. The approach velocity to the debris beds was 0.01 ft/s to be consistent with the strainers at STP.The results of this series of tests are summarized in the next section, and detailed results of thetests are presented after that.Summary of ResultsThe following conclusions can be drawn from this test series:" The fiber beds prepared with blending in a blender were not reproducible betweencolumns.
After 6 days of operation, the head loss varied from 1.2 inches of water to 61inches of water through debris beds that were circulating the same water at the same rate(see Figure 1).* The fiber beds prepared with the blended preparation method formed small, densenodules of fiber at the base of the fiber bed, immediately adjacent to the perforated support plate. The nodules formed a dimple pattern that matched the pattern of holes inDocument No: CHLE-O1O, Rev 2 Page 4 of 32Document No: CHLE-01 0, Rev 2Page 4 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Additionthe perforated plate, indicating that the smaller fibers formed by the blending processwere able to form a more dense fiber mat in a localized area (see Figure 14).* The difference in head loss among the three columns with the blended fiber preparation appears to be due to a trace amount of dirt or other material that collected in the small,dense fiber nodules at the base of the fiber bed. The debris bed with the highest head lossvisually had the greatest amount of darker material present in the nodules.
The fiber bedswere visibly clean through the rest of the depth, suggesting that little or no head lossoccurred through the bulk of the depth of the fiber bed and that nearly all of the head lossoccurred in the fiber where it contacted the perforated plate, indicating significant nonhomogeneity to the head loss characteristics of the bed (see Figure 14).* The fiber beds prepared with the NEI pressure-washing method were reproducible between columns.
After 6 days of operation, the head loss varied from 0.36 inches ofwater to 0.48 inches of water through debris beds that were circulating the same water atthe same rate (see Figure 2). Similar behavior continued until the test was terminated after 12 days.* The fiber beds prepared with the NEI pressure-washing method did not form the densenodules of fiber that were observed with the blended fiber beds (see Figure 15). Theabsence of these nodules and the reproducible behavior of the NEI beds lends furthercredibility to the conclusion that these nodules were responsible for the non-reproducible behavior of the blended fiber beds." The blended fiber preparation method resulted in shorter fibers (often called "shards" or"fragments")
than with the NEI pressure-washing method (see Figure 9 and Table 1),which may have led to the ability of the bed to form the dense nodules at the holes in theperforated plate. The shorter fibers led to a more compact debris bed (see Figure 10).Low porosity nodules are formed by local bed compaction, enabled by the mobility ofshort fiber shards formed during the chopping procedure.
Local compaction in regions offlow acceleration near the strainer penetrations is further enabled by loosely aggregated beds formed under very low approach velocity.
" The debris beds did not change thickness significantly over the course of the test, whichlasted over 8 days for the blended fiber debris bed and over 12 days for the NEI fiberdebris bed (see Figures 11 and 12). Further, minimal differences in thickness wereobserved between beds despite the wide variation in measured pressure loss. The lack ofchange in bed thickness, coupled with the visually clean nature of the debris beds, lendscredibility to the conclusion that the head loss associated with the blended fiber debrisbed was due to localized conditions at the perforated plate." Turbidity was an excellent indicator of the precipitation of aluminum hydroxide precipitates in solution (see Figure 22, 23, and 24).* Essentially all of the aluminum that was added to solution formed a precipitate, asindicated by turbidity measurement and supported by total and filtered aluminumanalyses.
" The addition of I mg/L of aluminum in solution caused the formation of in-situprecipitates that caused head loss through the blended fiber debris beds. An additional 5mg/L (6 mg/L total) caused sufficient head loss to terminate the test (see Figure 29).* Aluminum in solution in the form of pre-formed WCAP precipitates also caused headloss sufficient to terminate a similar blended fiber debris bed test (see Figure 30).Document No: CHLE-010, Rev 2Page 5 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition* The addition of up to 40 mg/L of aluminum over a period of 6 days was not sufficient tocause head loss through the NEI fiber debris beds (see Figure 31). However, the sameamount of aluminum in the form of pre-formed WCAP precipitates caused extensive head loss that terminated a similar test within 3 hours3.472222e-5 days <br />8.333333e-4 hours <br />4.960317e-6 weeks <br />1.1415e-6 months <br /> (see Figure 32). These resultsindicate that precipitates formed in-situ through the addition of aluminum nitrate at aslow rate have significantly different characteristics from those of the pre-formed WCAPprecipitates.
" Particle size analyses indicate that the size of precipitates formed in-situ are up to 10times smaller than the pre-formed WCAP precipitates with similar solution conditions (0.17 gim versus 1.6 gim in diameter, see Table 2). This significant difference in sizeappears to be sufficient to cause the pre-formed WCAP precipitates to be retained by theNEI fiber debris beds but allow the in-situ precipitates to pass through the NEI fiberdebris beds. The results indicate that head loss may be less significant than indicated bythe use of pre-formed WCAP precipitates, depending on the filtration characteristics ofthe debris bed.* Zeta potential analyses indicate that solution chemistry affects the surface charge of pre-formed WCAP precipitates.
When tests were conducted in deionized water, pre-formed WCAP precipitates had a positive zeta potential.
When tests were conducted inAlbuquerque tap water, the pre-formed WCAP precipitates were nearly neutral.
Whentests were conducted in water containing boric acid and TSP, the precipitates had asignificant negative charge. The reversal of charge depending on solution chemistry suggests that head loss testing using pre-formed WCAP precipitates may not adequately predict the extent of head loss through a strainer under conditions in which debris bedsare not reliant on physical sieving for the retention of particles.
- Despite the slow introduction of aluminum
- nitrate, corrosion conditions were notperfectly emulated.
It is possible that conditions that minimize aluminum
- release, such asthe passivation of aluminum surfaces or the formation of a low-solubility oxide layer,would result in an upper limit to the aluminum concentration measured in solution.
Inaddition, saturation conditions relevant for direct nucleation of precipitation productswithin the fiber bed may have been artificially exceeded.
If direct nucleation is a credibleformation mechanism, then the tests reported here best describe the filtration behaviorbetween the two preparation methods and not necessarily the in-situ head-loss sensitivity.
Direct nucleation would avoid the complications of particle filtration, perhaps leading todifferent head loss response in the debris beds. As a result, additional tests thatinvestigate precipitate formation under prototypical corrosion conditions are needed.The following sections of this report provide detailed results from the experiments.
Individual sections are presented on (1) head loss, (2) approach
- velocity, (3) temperature, (4) bed formation and morphology, (5) water chemistry, and (6) effect of aluminum addition.
Following thosesections, overall conclusions of this test series are presented.
Head Loss through Fiberglass Debris BedsThe primary diagnostic parameter monitored during these tests was the head loss through thefiberglass debris beds. The head loss through the two types of debris beds differed from eachother as displayed in Figures 1 and 2. For the blender-chopped bed, the head losses through theDocument No: CHLE-01 0, Rev 2Page 6 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Additionthree beds were relatively similar to each other during the initial period of operation.
The headlosses through beds 1, 2, and 3 were 0.6, 0.6, and 0.9 inches of water column after 90 minutes ofoperation, respectively, and 0.6, 0.6, and 1.0 inches of water after 5 hours5.787037e-5 days <br />0.00139 hours <br />8.267196e-6 weeks <br />1.9025e-6 months <br /> of operation, respectively.
As time progressed, the head losses through the beds deviated more. After 1 dayof operation, the head losses through beds 1, 2, and 3 were 0.7, 1.4, and 6.2 inches of water,respectively.
After 6 days of operation, the head losses through beds 1, 2, and 3 had stabilized at1.2, 21, and 61 inches of water, respectively, a 50-fold difference between beds 1 and 3.Head loss depends on characteristics of the debris bed and the fluid passing through the bed.Changing water temperature causes changes in fluid viscosity and density that changes themeasured head loss. To isolate changes in head loss due to changes in bed morphology independently of the changes in fluid viscosity and density, the head loss data can be corrected toa constant temperature.
The head loss data was corrected for viscosity by applying the ratio ofwater viscosity at the measured and standard temperatures, and corrected for density bycalculating the change in static head between pressure taps due to the decreasing fluid density athigher temperature.
In Figures 1 and 2, the head loss data have been corrected to a temperature of 104 'F (40 'C), which is the approximate temperature at the end of each of the tests.The debris beds prepared with the NEI pressure-washing protocol were more consistent with oneanother and more consistent over time. Column 3 started with less head loss than did the othercolumns, but after less than 1 day of operation, the head loss of column 3 had increased to besimilar to that of the other columns.
After 5 hours5.787037e-5 days <br />0.00139 hours <br />8.267196e-6 weeks <br />1.9025e-6 months <br /> of operation, the head losses through beds 1,2, and 3 were 0.64, 1.0, and 0.41 inches, respectively.
After 6 days of operation, the head lossesthrough beds 1, 2, and 3 were 0.42, 0.36, and 0.48 inches, respectively.
The similar behaviorbetween columns continued until the test was terminated after 12 days.10080Vo 60 -Column 2@3401-Column 30 ,ji " ' i i" i ..I I I'0 1 2 3 4 5 6 7 8 9Time (days)Figure 1: Temperature-corrected head loss through fiberglass debris beds prepared bychopping in a blender (corrected to 104 *F). Aluminum nitrate addition started at about6.75 days.Document No: CHLE-010, Rev 2 Page 7 of 32Document No: CHLE-01 0, Rev 2Page 7 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition2.0'LI-" L "- Column 1,ia,,,0* I0.0 , ""-- -"-H --F-"H ---
V I I0 1 2 3 4 5 6 7 8 9 10 11 12 13 14Time (days) -_______________________
Figure 2: Temperature-corrected head loss through fiberglass debris beds prepared usingthe NEI pressure-washing method (corrected to 104 °F).Approach Velocity through Fiberglass Debris BedsThe approach velocity was maintained near 0.01 ft/s in all three columns for both tests.Approach velocity was adjusted by throttling a valve on the discharge side of centrifugal pumpsthat fed each column independently of the others. The approach velocities are displayed inFigures 3 and 4. The throttle valves required occasional adjustment to maintain the desired 0.01ft/s approach velocity during the first several days of each test, but little or no adjustment wasrequired later in the tests. Overall, Figures 3 and 4 show that the velocity through the debris bedswas maintained at acceptable velocities over the duration of each test.Document No: CHLE-010, Rev 2Page 8 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition0.0400.030 4 -Column 10.0202 0.0100.000-Column 2-Column 30 1 2 3 4 5 6 7 89Time (days)Figure 3: Superficial filtration velocity through fiberglass debris beds prepared using theblending method.0.04.0.0304' 0.02.0.01-Column 1-Column 2-Column 30.00.....................~ i i i i I I T I I [ I l I I I ,0 1 2 3 4 5 6 7 8 9 10 11 1213 14Time (days)Figure 4: Superficial filtration velocity through fiberglass debris beds prepared using theNEI pressure-washing method.Document No: CHLE-010, Rev 2Page 9 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum AdditionTemperature Temperature Profile over TimeThe temperature during each test was designed to decline over time to reflect the temperature present in the containment building following a LOCA. The temperature in these tests wasmodeled to emulate a break of a 6-inch pipe in containment at STP (a medium break LOCA).The target temperature profile was generated by running the MELCOR/RELAP-5 computersimulations at Texas A&M University.
The predicted temperature started at about 185 'F (85°C) and declined rapidly in the first few minutes as the leaked water came in contact withconcrete and other surfaces.
The temperature then increased over several hours as materials within the containment building heated up, reaching a temperature of 162 'F (72 °C) about 4hours into the event. The CHLE temperature control system was designed to provide a moregradual decline to this temperature after 4 hours4.62963e-5 days <br />0.00111 hours <br />6.613757e-6 weeks <br />1.522e-6 months <br />, and then to track the temperature predicted byMELCOR/RELAP-5 for the remainder of the experiment.
The temperature profiles are shown inFigures 5 and 6.During the test with the fiber beds prepared by the blending method, the temperature droppedrapidly to 162 'F (72 'C) after 4 hours4.62963e-5 days <br />0.00111 hours <br />6.613757e-6 weeks <br />1.522e-6 months <br /> into the test, but was followed by a period of constanttemperature because of the failure of a temperature controller to initialize overnight.
Thetemperature was reduced to be in compliance with the desired temperature profile before the endof the first day, and remained within 5 'F (2.8 'C) of the desired temperature until the final dayof the test. The measured temperature exhibited a sawtooth pattern because the temperature controller tended to overshoot the target temperature.
The increase in temperature ranged from1.3 to 1.6 'F (0.7 to 0.9 'C) between a low-temperature reading and the next high-temperature point. Toward the end of the test, an attempt was made to adjust the deadband on thetemperature controller, but the amount of overshooting increased slightly.
The operation of the temperature controller was improved for the test with the NEI pressure-washed fiber, as presented in Figure 6. The measured temperature was consistently within about1.8 'F (1 'C) of the target temperature throughout the duration of the test and without theovershooting of the controller that had been exhibited in the test with the blended fiber.Temperature Variation in TankThe temperature was measured at four locations within the CHLE tank: three in the pool and onein the vapor space. The three locations in the pool consisted of a point near the center and twopoints in comers of the tank. The average temperature in comers was within 0.2 'F (0.1 'C) ofthe temperature in the center of the tank, indicating that the thermal conditions within the tankwere uniform.
The comparison between the temperature in the vapor space and in the center ofthe pool is shown in Figure 7. In general, the temperature in the vapor space was 4 to 5 'F (2.2to 2.8 'C) lower than the temperature in the center of the pool.Temperature Differential in Head Loss ColumnsThe temperature was also measured continuously in each head loss column about 6 inches abovethe debris screen. Initially, the temperature in the columns was somewhat lower than theDocument No: CHLE-010, Rev 2Page 10 of 320 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Additiontemperature in the tank because of heat loss in the connecting piping and in the columns.
Thedifference in temperature between the center of the tank and column 1 is shown in Figure 8. Theother columns are not shown because their results are virtually identical to column 1. At thebeginning of the test, the temperature in the columns was about 1.8 OF (1 °C) lower than in thetank. But as the tank temperature decreased, the temperature difference between the tank andcolumns decreased.
After 6 days, when the temperature in the tank had dropped to about 113 IF(45 'C), the temperature in the columns had converged to that in the tank.80U-70060I,-40300 1 2 3 4 5 6 7 8 9Time (days)Figure 5: Tank temperature during the experiment using fiberglass debris beds preparedusing the blending method.80L) 70a605040300 1 2 3 4 5 6 7 8 9 10 11 12 13 14Time (days)Figure 6: Tank temperature during the experiment using fiberglass debris beds preparedusing the NEI pressure-washing method.Document No: CHLE-010, Rev 2Page 11 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition10S8Eo ias0 L.E.00. 2-1--imp 0400 1 2 34 5 6 7 8Time (days)-i I 'i 11 1! , I i7 11 li, 11 19 1011 12 13 1411 12 13 14Figure 7: Difference between the pool temperature and the vapor-space temperature in thetank during the experiment using fiberglass debris beds prepared using the NEI pressure-washing method.E00*4)ICL C1.21.00.80.60.40.20.0-0.20 1 2 3 4 5 6 7 8Time (days)9 10 11 12 13 14Figure 8: Difference between the pool temperature and the temperature in Column 1 justabove the debris bed during the experiment using fiberglass debris beds prepared using theNEI pressure-washing method.Document No: CHLE-010, Rev 2 Page 12 of 32Document No: CHLE-010, Rev 2Page 12 of 32I Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum AdditionBed Formation and Morphology The difference in fiber preparation resulted in variability in the characteristics of the fibers andhow they formed debris beds. The difference in fiber preparation was evident when the debriswas examined on a light table, as shown in Figure 9. The photographs in Figure 9 show thecontents of a 8 x 13 inch glass pan. The blended fiber preparation consisted almost entirely ofindividual fibers with few or no clumps. The NEI fiber preparation contained more loose clumpsof fibers. In addition, the blended fiber method resulted in visually shorter fiber lengths.
Thevisual difference in fiber length was corroborated with data on fiber length generated by IPSTesting Services, a company that measures fiber length for the paper industry
[4]. IPS testedsamples of fiber prepared by the two methods and reported the data in Table 1. Sixty-two percent of the blended fibers were less than 0.5 mm, but only 44% of the NEI fibers were. Themeasurement system could not detect the length of fibers that were less than 0.2 mm, whichincluded 21 percent of the blended fibers and 13 percent of the fibers from the NEI preparation.
IPS reported NEI fibers up to 7 mm in length, but only a maximum of 3 mm for the blendedfibers.A BFigure 9: Examination of debris on a light table from (A) blended fiber preparation, and(B) NEI pressure-washed fiber preparation.
Table 1: Fiber Length of the Debris Prepared by Blended and NEI Preparation MethodsBlended preparation NEI pressure-washed preparation
(% of fibers) (% of fibers)< 0.5 mm 62 440.51 to 1.0 mm 18 24> 1.0 mm 13 32The difference in fiber preparation was also evident when the debris was placed in the head losscolumns, as shown in Figure 10. The beds were formed with the same approach velocity in bothtests, 0.1 ft/s. The blended fiber preparation shown in Figure 1 0A had an initial bed thickness ofabout 0.6 inches (1.5 cm), but the NEI pressure-washed fiber preparation shown in Figure lOBhad an initial overall bed thickness of about 2.6 inches (6.5 cm), even though both types of fiberbeds contained the same amount of fiber on a mass basis (18 grams). These thicknesses Document No: CHLE-010, Rev 2Page 13 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Additioncorrespond to an in-place initial density of 4.0 lb/ft3 for the blended fiber beds and 0.9 lb/ft3 forthe pressure-washed fiber beds.For each type of fiber preparation, the beds in the three head loss columns were fairly consistent with each other. For the blended fiber preparation, the thickness of the bed in each column wasslightly different, with Column 1 having the thickest bed and Column 3 having the thinnest bed.The trend in thickness is consistent with the trend in accumulated head loss at the end of the test(i.e., Column 3 had the thinnest bed and the highest head loss). However, the differences wereminor, as shown in Figure 11. Furthermore, the thickness of the debris beds did not changesignificantly over the duration of the test, even though the head loss changed dramatically forColumns 2 and 3, as noted earlier.
The change in head loss over the duration of the tests cannotbe attributed to the change in bed thickness, given that head loss is expected to vary linearly withbed thickness for a perfectly homogeneous debris configuration.
Similarly, the debris beds formed of NEI pressure-washed debris in the three columns weresimilar in overall thickness, but Column 2 had the thickest bed, followed by Columns I and 3, asshown in Figure 12. The top surface of the NEI pressure-washed debris beds was less uniformthan for the blended fiber beds. The NEI pressure-washed debris beds were thicker in the centerof the column and somewhat thinner at the edges where the debris bed touched the column wall.A BFigure 10: Debris beds in head loss columns at the beginning of test operation.
(A)Column 2 blended fiber debris bed and (B) Column 2 NEI pressure-washed fiber debrisbed.Document No: CHLE-010, Rev 2Page 14 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition2.01.00.5**Column 1IColumn 2A Column 3.....I0.0012345678Time (days)Figure 11: Thickness of blended fiber debris beds in head loss columns.4.03.53.0UI-,-5.2.5eW 2.0IJ1.51.0I*Column IIColumn 2A Column 30.08010024612lime (days)Figure 12: Thickness of NEI pressure-washed fiber debris beds in head loss columns.Differences were also observed when the tests were complete and the debris beds were removedfrom the head loss columns.
The blended fiber debris beds compressed slightly when water wasremoved from the column, from a thickness of about 0.6 inches (1.5 cm) to 0.4 inches (1 cm).After water was removed, water droplets falling from the inside of the column formed craters onthe otherwise uniform surface of the blended fiber beds, as shown in Figure 13, indicating thatthe surface of the bed was soft and pliant.As water was drained from the columns, the NEI pressure-washed debris beds collapsed moresignificantly, from about 2.6 inches (6.6 cm) to 0.8 inches (2 cm) in thickness.
The surface ofthe NEI debris beds was much rougher and less uniform after being removed from the column,as shown in Figure 13C.Document No: CHLE-01 0, Rev 2Page 15 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum AdditionACFigure 13: Photographs of the debris beds after the tests were complete.
(A) Column 2blended fiber debris bed in column after water was drained and clear section was removedfrom the test assembly (craters are from falling water droplets).
(B) Column 2 blendedfiber debris bed after being removed from column. (C) Column 1 NEI pressure-washed debris bed after being removed from column.After being removed from the head loss columns, the debris beds were stored in a darkrefrigerated room at 4 'C until they could be examined in detail. For the blended debris beds,visual examination indicated the blended fiber beds to have a uniform, flat top surface.
Thecraters from the water droplets in the bed from Column 3 appeared to have filled in over timeand were less prominent than they had been when the bed was removed from the column. In allbeds, the fibers were tightly packed so that the surface of the bed appeared similar to a piece offelt material.
Individual fibers were not readily visible.
Examination with a 16x magnification lens (Loupe) did not provide a significantly different view of the bed surfaces.
When a portion of the bed was torn from the surface and rubbed between two fingers, it stayed asan intact mat of fibers. However, the debris beds were easily torn and exhibited very littleresistance when pulled apart by hand. The debris beds were cut vertically with a scissors, andthe cross-section of the bed appeared uniform from top to bottom, also visibly similar to a pieceof felt.The most significant feature of the blended debris beds appeared at the bottom of the bed, wherethe debris bed was in contact with the perforated support plate. When the debris beds werepeeled from the support plate, they exhibited nodules of fiber in a dimple pattern identical to theholes in the support plate. The bottom of the bed of all three columns with blended fiberpreparation are shown in Figure 14. The texture of the nodules was significantly different fromthat of the fiber at the top or center of the debris bed. When rubbed between two fingers, thenodules were hard, almost as if a rock were embedded in the debris bed. When a nodule waspulled apart by hand, however, it offered no physical resistance and separated easily into a smallmass of fibers with no evident larger solid materials.
- Visually, the nodules in bed 3 were darkerthan in bed 2, which were subsequently darker than in bed 1. The color might indicate thatanother material such as dirt was retained in the localized area of the fiber immediately at theperforated screen holding the debris beds. The debris beds appeared clean throughout theirdepth, as evidenced in the cross-section of the beds shown in Figure 14. The difference in theretention of dirt in the fiber at the perforated plate may have been the cause of the difference inDocument No: CHLE-O1O, Rev 2 Page 16 of 32Document No: CHLE-01 0, Rev 2Page 16 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Additionhead loss among the three columns, being that the trend in color is consistent with the trend inhead loss (darker color corresponding to greater head loss).The dimple pattern evident at the bottom of the blended beds was barely evident on the NEIpressure-washed beds, as shown in Figure 15. These images indicate that the longer fiberspresent in the pressure-washed beds spanned the holes of the perforated screen differently, leading to different hydraulic conditions in the immediate vicinity of the support screen. Thedifference in fiber bed characteristics at the support screens appears to be a significant cause ofthe difference in head loss among the three columns for the blended fiber beds and between theblended fiber and pressure-washed columns.A BFigure 14: Dimple pattern from the support plate on the bottom side of the blended fiberdebris beds in (A) Column 1, (B) Column 2, and (C) Column 3. The dimples in Column 3are darker than in Column 1. Also, the cross section of the debris bed, most evident onColumn 3, appears clean throughout the entire depth of the debris bed.ABCFigure 15: Bottom side of the NEI pressure-washed fiber debris beds in (A) Column 1, (B)Column 2, and (C) Column 3.Document No: CHLE-0 10, Rev 2 Page 17 of 32Document No: CHLE-010, Rev 2Page 17 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum AdditionWater Chemistry pHThe pH in the tank solution was measured using both an on-line pH probe and grab samplesmeasured with a tabletop laboratory pH meter. The on-line probe was more difficult to calibrate than the laboratory pH meter and could not be calibrated while a test was in progress whereas thelaboratory pH meter could be calibrated.
The pH measurements by both the in-line probe andthe laboratory pH meter are shown in Figure 16 for the blended fiber test and Figure 17 for theNEI fiber test. During the blended fiber test, the average pH was 7.35 using the on-line pHprobe and 7.18 using the laboratory pH meter. The laboratory pH meter was calibrated each timeit was used and was closer to the expected pH value of 7.2 based on calculations of the chemicals in the solution.
The on-line probe was replaced with a new probe before the NEI test started.For the NEI fiber test, the average pH value was 7.13 using the on-line probe and 7.23 using thelaboratory pH meter. Again, the laboratory pH meter was closer to the expected pH value of 7.2.Calcium and SilicaCalcium and silica were measured in the pool solution, along with the aluminum.
The calciumresults are shown in Figure 18 for the blended fiber beds and Figure 19 for the NEI fiber beds.The calcium concentration was fairly constant with time in both tests. In the blended fiber test,the calcium concentration was about 1.5 mg/L but in the NEI fiber test was somewhat higher,about 2.0 to 2.2 mg/L.The silica results are shown in Figure 20 for the blended fiber beds and Figure 21 for the NEIfiber beds. The results were similar in both tests, starting at about 2.5 mg/L (after the TSP hadbeen added to the tank and the solution had been circulating through the head loss columns forabout 90 minutes) and gradually rising over a period of several days to a concentration of about4.0 mg/L, where it stayed for the remainder of the tests.9.0 -- On-line pH8,5 _ Bench PH8.06.50 1 2 3 4 5 6 7 8 9Time (days)Figure 16: pH during the blended fiber test.Document No: CHLE-010, Rev 2Page 18 of 32I Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition9.08.58.07.57.06.56.00 1 2 3 4 5 6 7Time (days'8 9 10 11 12 13 14Figure 17:pH during the NEI fiber test.5-4E 3E20#Total Calcium (mg/L)U Filtered Calcium (mg/L)a i 'a ti "0 12 3 4 5Time (days)6 7 89Figure 18:Measured calcium in the CHLE pool solution during the blended fiber test.Document No: CHLE-010, Rev 2Page 19 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition54E3E201-0#Total Calcium (mg/L)0 Filtered Calcium (mg/L)Uai ..I ; ....i .I I .i ...I ! ..I .! I ý ý I .ý ..02468101214Time (days)IFigure 19: Measured calcium in the CHLE pool solution during the NEI fiber test.542(U*.2 2Coi*Total Silica (mg/L)U Filtered Silica (mg/L)a a00 1 23 4 5 6 78 9Time (days)Figure 20: Measured silica in the CHLE pool solution during the blended fiber test.54-03E.2 2* I a ~N a0 1U)1 -~ *Total Silica (mg/L)U Filtered Silica (mg/L)0 2 4 6Te3 8sst Time (days)101214Figure 21: Measured silica in the CHLE pool solution during the NEI fiber test.Document No: CHLE-010, Rev 2Page 20 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum AdditionEffect of Aluminum AdditionAfter 6.7 days of operation with no materials in the tank, aluminum nitrate were added to thetank in each test. Aluminum nitrate was added at a slow rate (an increase of 0.02 mg/L perminute) to minimize the potential for precipitation to occur due to localized high concentration atthe point of injection.
For the blender preparation method, aluminum addition continued until aconcentration of 6 mg/L as Al was reached in the solution with, and the test was terminated whenthe head loss in two of the three columns exceeded 80 inches of water. No significant change inhead loss was observed in the test with the NEI pressure-washed preparation method; therefore, the aluminum addition was added in 7 periodic batches until 40 mg/L as Al was reached in thesolution.
Turbidity Several parameters were monitored as aluminum was added, including
Determining theconcentrations and suspended solids required sample analysis that took several days, butturbidity can be determined immediately.
Turbidity correlated well with the formation ofprecipitates in the solution.
In the test with the blended fiber, the turbidity started at about 0.81NTU after the TSP was added and the CHLE tank was circulating through the head losscolumns.
Turbidity dropped slowly over several days, reaching 0.32 NTU on Day 6. Afteraluminum was added, the turbidity increased in direct proportion to the quantity of aluminumthat was added, as shown in Figure 22. Turbidity is determined by measuring scattered light in asolution and is an indicator of particles in solution.
Therefore, the turbidity is evidence that aprecipitate was forming.
Furthermore, the slight decrease in turbidity from 0.97 NTU after thefirst addition of aluminum was complete to 0.78 NTU before the second addition started(between days 7 and 8) may be an indication of removal of precipitates, possibly by beingcaptured in the debris beds.Similar results were observed for the test with the NEI fiber preparation.
The turbidity started atabout 0.63 NTU after the TSP was added and the CHLE tank was circulating through the headloss columns.
The turbidity then dropped slowly, reaching 0.40 NTU on Day 6. After aluminumwas added, the turbidity again increased in direct proportion to the quantity of aluminum thatwas added, as shown in Figure 23. The strength of this correlation indicates that essentially allof the aluminum that was being added was precipitating as particles in solution.
The correlation between the amount of aluminum added and the turbidity in the solution for the NEI preparation method is shown in Figure 24.Aluminum Concentration Samples were taken for total and filtered aluminum and analyzed by a commercial laboratory.
The results of the analysis are shown in Figure 25 for the blended fiber beds and Figure 26 forthe NEI fiber beds. In both cases, the total aluminum in solution agrees well with the amount ofaluminum that was added in solution.
- However, the concentration of aluminum in the filteredsamples is significantly lower. These results indicate that a portion of the aluminum wasremoved during the filtration
- process, as would be expected when precipitates form. However,in both cases, several mg/L of aluminum were measured in the filtered
- samples, indicating thatnot all of the precipitates that were detected by the turbidity measurements were removed by theDocument No: CHLE-010, Rev 2Page 21 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Additionlaboratory filter. The standard filter for measuring filtered metals, however, has a nominal poresize of 0.45 gm. As will be shown later, evidence was found that the in-situ precipitates formedby the addition of aluminum nitrate were smaller than this size and a portion of the precipitates could have passed through the filter.76'420654:5302 E0 2 4 6 8Time (days)010Figure 22: Measured turbidity in solution, and aluminum added to the CHLE tank overtime during test with blended fiber preparation.
454035E 30"W 25E 20E15R 105030252015 V5105002 4 6 8 10 12Time (days)14Figure 23: Measured turbidity in solution, and aluminum added to the CHLE tank overtime during test with NEI pressure-washed fiber preparation.
Document No: CHLE-010, Rev 2 Page 22 of 32Document No: CHLE-010, Rev 2Page 22 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition3025m 20I-z1510500 5 10 15 20 25Aluminum Added (mglL)30 35 40 45Figure 24: Correlation between measured turbidity in solution and amount of aluminumadded to the CHLE tank in the form of aluminum nitrate.76z5-4Ec 310-Aluminum addition* Total Aluminum (mg/L)-- -" Filtered Aluminum (mg/L)i6.06.57.07.58.08.5Time (days)Figure 25: Measured total and filtered aluminum concentration in the CHLE pool solutionduring the blended fiber test, along with the aluminum that was added.Document No: CHLE-010, Rev 2Page 23 of 32I Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition50-4030E"E 2010-Aluminum addition* Total Aluminum (mg/L){ .Filtered Aluminum (mg/L)21o.0i0246810Time (days)12 14CHLE pool solutionFigure 26: Measured total and filtered aluminum concentration in theduring the NEI fiber test, along with the aluminum that was added.Total Suspended SolidsTotal suspended solids (TSS) was a less reliable indicator of the presence of precipitates thanturbidity was. The TSS results are shown in Figure 27 for the blended fiber beds and Figure 28for the NEI fiber beds. The factor most likely contributing to the poor correlation between TSSand precipitation formation was the particle size of the precipitates relative to the nominal poresize of the filter. The filter used for the TSS analysis is a glass-fiber filter with a nominal poresize of 1.2 jLrm (Whatman GF/C). As will be demonstrated later, the precipitates had an averagesize smaller than that, and many of the particles may have passed through the filter withouthaving been measured as suspended solids. Prior to the addition of aluminum
- nitrate, the TSS ofthe solution was between 3 and 5 mg/L for both tests.14I1412.10._jIm8E; 6420765E=1J--+0lO02 4 6 8Time (days)2 01'0Figure 27: Measured total suspended solids (TSS) in the CHLE pool solution during theblended fiber test, along with the aluminum that was added.Document No: CHLE-010, Rev 2Page 24 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition16 4514
- TSS mg/L 403512 -Aluminum addition 25" 210 * "t *25 EE*nf* 20 .=*_ E2)215 .4*00 2 4 6 8 10 12 14Time (days)Figure 28: Measured total suspended solids (TSS) in the CHLE pool solution during theNEI fiber test, along with the aluminum that was added.Debris Bed Head LossThe head loss trends after the aluminum was added to the blended fiber beds are shown in Figure29. After 1 mg/L of aluminum had been added, the head loss in all three columns startedincreasing.
After 12 hours1.388889e-4 days <br />0.00333 hours <br />1.984127e-5 weeks <br />4.566e-6 months <br /> of operation, Column 3 reached its maximum head loss and itsoperation was terminated.
The following day, 5 mg/L of aluminum was added (yielding a totalof 6 mg/L). The head loss in Columns 1 and 2 started increasing
- rapidly, and the test wasterminated after an additional 6 hours6.944444e-5 days <br />0.00167 hours <br />9.920635e-6 weeks <br />2.283e-6 months <br />. Details of the statistical significance of the change inhead loss after aluminum was added is described in Appendix 1.Previous tests were conducted by adding pre-formed WCAP precipitate directly to the CHLEtank in batches.
These tests were previously summarized in CHLE-008.
The results of a testwith WCAP precipitate and blended fiber beds are shown in Figure 30. The tests with in-situprecipitate formation and WCAP precipitate addition are difficult to compare because the latterhad larger quantities of precipitate added over shorter periods of time. In addition, the tests inthis series had been exposed to circulating solution for over 6 days, which may have led tosufficient local compaction to permit a more rapid response to the aluminum nitrate compared tothe earlier tests. In the WCAP tests, five batches of WCAP precipitate, totaling 21.8 mg/L ofaluminum, were added over a 4-hour period. Head loss started increasing rapidly about 4 hour4.62963e-5 days <br />0.00111 hours <br />6.613757e-6 weeks <br />1.522e-6 months <br />safter the first addition of WCAP precipitate, and the test was terminated an hour later. However,it is evident that in the case of blended fiber beds, both in-situ precipitate formation and theaddition of pre-formed WCAP precipitates can cause significant head loss. It should be notedthat the timing of head loss generation can be influenced by physical arrangement of debris in thebed and the degree of compaction that has been allowed to occur prior to the chemical arrival.Document No: CHLE-010, Rev 2Page 25 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition120-Column 1 1 mg/LAl added 5 mg/LAl added100 -_Column 2
36000 60_tS40-20 -"140 150 160 170 180 190 200Time of operation (hours)Figure 29: Head loss through debris beds with blended fiber preparation when in-situprecipitation occurred due to the addition of aluminum nitrate to the CHLE tank.8070S60.9- 5040"u 40_ 30.2201000 1 2 3 4 5 6 7Time of operation (hours)8Figure 30: Head loss through debris beds with blended fiber preparation when pre-formed WCAP precipitates were added to the CHLE tank.Significantly different results were obtained when in-situ precipitation was achieved in the testswith NEI pressure-washed fiber preparation.
The initial 1 mg/L and 5 mg/L aluminum additions Document No: CHLE-010, Rev 2Page 26 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Additiondid not cause any change in head loss. Thus, head loss additions continued daily for severaladditional days, until a total of 40 mg/L of aluminum had been added to the solution.
Thealuminum addition and head loss results are shown in Figure 31. Column 2 had a change inperformance after 16 mg/L of aluminum had been added, and this change in performance wasdetermined to be statistically significant, as described in Appendix
- 1. However, the changeappears to be in the range of 0.1 inches of water, a change in head loss that would not beconsidered problematic during a LOCA. No statistically significant changes in head lossoccurred in Columns 1 and 3 even after 40 mg/L of aluminum had been added.Figure 32 shows that the NEI pressure-washed fiber beds behaved significantly differently whenpre-formed WCAP precipitates were added to the CHLE tank. The same quantity of aluminum(40 mg/L) was added in the form of WCAP precipitates, except that it was added in a muchshorter period of time (about 2 hours2.314815e-5 days <br />5.555556e-4 hours <br />3.306878e-6 weeks <br />7.61e-7 months <br /> for WCAP versus about 5 days for the in-situ precipitate formation) and the WCAP was added in three batches versus seven addition cycles for the in-situprecipitate formation.
- However, when WCAP precipitates were added, the debris beds rapidlyreached their maximum head loss values. When the same amount of aluminum was added tocause in-situ precipitation, no head loss occurred over a period of days, despite that the sametype of debris beds (NEI pressure-washed) were present, and with the same quantity of fiber (18g per bed). A comparison of Figures 31 and 32 is evidence that the slow formation ofprecipitates in-situ due to the addition of aluminum nitrate over a period of days into a buffered, borated solution typical of a LOCA at STP is not the same as the precipitates that are formed bythe WCAP protocol.
The same addition of aluminum caused over 80 inches of water of headloss when added as the WCAP precipitate, versus less than 1 inch of water of head loss whenformed as an in-situ precipitate.
It is important to understand the physical mechanisms thatpermit NEI fiber to pass in situ chemical products compared to the significant filtration ofWCAP material.
2.0 50-Column 1 40-Column 2 _'j41.5 Column 3 Eo [Al added 200i1.0t -10*00S0.5 ..I!0.0140W.M I34340190 240 290Time of operation (hours)Figure 31: Head loss through debris beds with NEI pressure-washed fiber preparation when in-situ precipitation occurred due to the addition of aluminum nitrate to the CHLEtank.Document No: CHLE-010, Rev 2Page 27 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition--=-Column 1-670 -Column 2 40A60 -Al Added CM50 +20uo40 10 -Vo.E_ '-itw 30j04o 20 -6Eto dP cell changed ! -14 15 16 17 18 19Time of operation (hours)Figure 32: Head loss through debris beds with NEI pressure-washed fiber preparation when pre-formed WCAP precipitates were added to the CHLE tank.Particle Size and Zeta Potential To determine the cause of the difference in performance of the in-situ and pre-formed WCAPprecipitates, the particle size and zeta potential of the precipitates were measured.
The zetapotential is a measure of the surface charge of particles, which can affect their ability toaggregate into larger particles or to be retained in a filter. The analysis revealed distinctdifferences between the in-situ and pre-formed WCAP precipitates.
The particle size and zetapotential measurements are summarized in Table 2. The preformed WCAP precipitates had aparticle size of about 1.6 ptm when they were in a solution containing boric acid and TSP at theconcentrations used in the CHLE tests. This size was consistent over three samples of WCAPprecipitates.
When the WCAP precipitates were placed in a solution containing water that hadbeen deionized by reverse osmosis, the particle size was slightly larger, ranging from 1.8 to 2.5gm. This difference in size is unlikely to make a substantive difference in how the particles behave in solution or their ability to be retained in a debris bed. When the WCAP precipitates were placed in a solution containing tap water, the particle size was about the same size, 1.2 ptm.An important result in Table 2 is the size of the in-situ precipitates.
These precipitates weremeasured at 0.17 gm on the day that the NEI test was completed.
This size is one-tenth that ofthe pre-formed WCAP precipitates.
The substantial difference in size may help explain why thepre-formed WCAP precipitates were retained by the NEI fiber beds but the precipitates formedin-situ were not. Samples of solution from the NEI test were stored in a laboratory oven at thefinal temperature of the test (40 °C) for two weeks, after which the particle size was measuredagain. The particle size was virtually identical in the second measurement, indicating that noaggregation had taken place over the storage period. In addition, visual observation indicated that no settling had occurred over the storage period, which is consistent with the small size ofthe particles.
Document No: CHLE-010, Rev 2Page 28 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum AdditionTable 2: Particle size and zeta potential of aluminum hydroxide precipitates Particle Size (jtm) Zeta Potential (mV)Pre-formed WCAP Precipitates In boron/TSP water, 30 mg/L dilution 1.6 -28In boron/TSP water, 40 mg/L dilution 1.6 -25In boron/TSP water, 10 mg/L dilution 1.6 -27In deionized water, 40 mg/L dilution 1.8 29In deionized water, 10 mg/L dilution 2.5 30In tap water 1.2 -3In-situ Precipitates (in boron/TSP water)At end of test 0.17 -31Two weeks later 0.18 -321. Each value given is the average of three measurements When precipitates were placed in water containing boric acid and TSP, they had a zeta potential ranging from -25 to -32, regardless of whether they were pre-formed WCAP precipitates or in-situ precipitates.
The in-situ precipitates were slightly more negative than the pre-formed WCAPprecipitates.
When particles are retained in a debris bed entirely by mechanical
- sieving, thecharge on the particles is not important.
- However, when particles are smaller than the voiddimensions in the debris bed, surface charge can have a significant impact on particle retention.
Particles that have similar charge to the media in the debris bed will have electrostatic repulsion with the media and are less likely to be retained.
Particles that are neutral or have oppositecharge to the media can be retained by van der Waals forces or by electrostatic attractive forces.Previous work by Duke Energy indicated that Nukon fibers are also negatively
- charged, althoughthe magnitude of the charge depended on the solution in which the fibers were immersed.
At apH of about 7, that report found that Nukon fibers had a zeta potential of about -25 in watercontaining boric acid but about -12 in either deionized water or tap water. The report by DukeEnergy noted that head loss through fiber beds could be a function of the type of water used, andattributed some of the difference in head loss performance to the zeta potential.
Because the in-situ particles and the Nukon fiber both have significant negative charge whenimmersed in water containing boric acid and TSP, they will have repulsive electrostatic forcesand will not be retained unless the void dimensions in the debris bed are small enough tophysically sieve the particles.
These effects may explain the differences between the blendedand NEI debris beds. The blended beds had hard nodules of fiber adjacent to the perforated screen, and these nodules had apparently retained some dirt or other material, leading to adifference in head loss among the three columns.
The nodules apparently had small enough voiddimensions that they were able to retain both pre-formed WCAP precipitates and in-situprecipitates.
The NEI fiber bed, however, did not have the hard nodules of fiber and had voidspaces small enough to retain the pre-formed WCAP precipitates but not small enough to retainthe in-situ precipitates, which were one-tenth the size.Document No: CHLE-010, Rev 2Page 29 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum AdditionThe results in Table 2 also indicate that the zeta potential of pre-formed WCAP precipitates wasdependent on solution chemistry.
In water containing boric acid and TSP, the precipitates werenegatively charged with a zeta potential around -27. In tap water, the precipitates were nearlyneutral with a zeta potential around -3. In deionized water, the precipitates were positively charged with a zeta potential around +30. These results could be significant for the use of pre-formed WCAP precipitates for some head loss tests. If debris beds are formed that are porousenough that they do not physically sieve precipitates, pre-formed WCAP precipitates that areused in tests containing deionized water or tap water are likely to be retained in the debris bed byelectrostatic attraction or van der Waals attractive forces. In contrast, if the tests used watercontaining boric acid and TSP, the precipitates would be less likely to be retained because of therepulsive electrostatic forces, and less head loss would occur. Neither deionized nor tap waterconditions exist within the post LOCA accident environment, so negative repulsive forces arelikely to exist between fiberglass debris and aluminum based chemical products that may form inthe sump environment.
Conclusions The tests described in this report were developed to investigate (1) the suitability of debris bedsfor use in long-term corrosion/head loss experiments, which may be up to 30 days long, and (2)the response of the debris beds to the slow addition of aluminum
- nitrate, which can simulate therelease of aluminum into solution by corrosion.
The fiber beds prepared with blending in a blender were not reproducible between columns.After 6 days of operation, the head loss varied from 1.2 inches of water to 61 inches of waterthrough debris beds that were circulating the same water at the same rate. The lack ofreproducibility would make it difficult to use this fiber bed preparation method in the CHLEprogram to assess the importance of chemical
- effects, because the difference in performance ofdue to variability may be greater than the difference in performance due to the addition ofchemical effects.
The lack of reproducibility of the fiber beds with the blended preparation appears to be due to localized regions of more dense fiber packing immediately adjacent-to theperforated support plate.In contrast, the fiber beds prepared with the NEI pressure-washing method were reproducible between columns and had steady performance over time. Testing by an outside laboratory demonstrated that the NEI pressure-washing method results in longer fibers. The shorter"shards" of fibers in the blended beds may be a key reason for the irreproducibility of thosedebris beds.When aluminum nitrate was introduced into the solution, precipitates formed, as evidenced byturbidity measurements.
There was a strong correlation between the turbidity measurements andthe amount of aluminum that had been added. After aluminum nitrate was introduced into thesolution, the head loss through the blended fiber beds increased.
For the NEI pressure-washed debris beds, one of the three beds had a statistically significant response to the addition ofaluminum
- nitrate, but the change in head loss was only about 0.1 inches of water in that column,which is not enough to cause concern during a LOCA. The other two NEI pressure-washed Document No: CHLE-010, Rev 2Page 30 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Additiondebris beds did not have a statistically significant response to the addition of aluminum nitrate,even though in earlier tests they did experience a significant increase in head loss after pre-formed WCAP precipitates were added to the system.Particle size analyses indicate that the size of precipitates formed in-situ are up to 10 timessmaller than the pre-formed WCAP precipitates with similar solution conditions (0.17 jim versus1.6 gtm in diameter).
This significant difference in size appears to be sufficient to cause the pre-formed WCAP precipitates to be retained by the NEI fiber debris beds but allow the in-situprecipitates to pass through the NEI fiber debris beds. The results indicate that head loss may beless significant than indicated by the use of pre-formed WCAP precipitates, depending on thefiltration characteristics of the debris bed.Zeta potential analyses indicate that solution chemistry affects the surface charge of pre-formed WCAP precipitates.
When tests were conducted in deionized water, pre-formed WCAPprecipitates had a positive zeta potential.
When tests were conducted in Albuquerque tap water,the pre-formed WCAP precipitates were nearly neutral.
When tests were conducted in watercontaining boric acid and TSP, the precipitates had a significant negative charge. The reversal ofcharge depending on solution chemistry suggests that head loss testing using pre-formed WCAPprecipitates may not adequately predict the extent of head loss through a strainer underconditions in which debris beds are not reliant on physical sieving for the retention of particles.
Despite the slow introduction of aluminum
- nitrate, corrosion conditions may not have beenperfectly emulated in these tests. It is possible that conditions that minimize aluminum release,such as the passivation of aluminum surfaces or the formation of a low-solubility oxide layer,would result in an upper limit to the aluminum concentration measured in solution.
In addition, saturation conditions relevant for direct nucleation of precipitation products within the fiber bedmay have been artificially exceeded.
If direct nucleation is a credible formation mechanism, then the tests reported here best describe the filtration behavior between the two preparation methods and not necessarily the in-situ head-loss sensitivity.
Direct nucleation would avoid thecomplications of particle filtration, perhaps leading to different head loss response in the NEIpressure-washed debris beds. As a result, additional tests that investigate precipitate formation under prototypical corrosion conditions are needed.It is important to note that a growing suite of routine diagnostics is available to detect andcharacterize the presence of chemical products.
These tests include:
optical turbidity, analytical measurements of solution chemistry, particle sizing, zeta potential, SEM fiber examination, chemical addition mass balance, online differential
- pressure, and visual estimation of debrisbeds. There is no exclusive dependency on differential pressure to detect the presence ofchemical products.
The use of several of these experimental methods can be used to detect thepresence of chemical precipitates in an integrated corrosion/head loss test to evaluate whetherprecipitates that may form following a corrosion process are similar to the precipitates detectedhere after the addition of aluminum nitrate.
In addition, the tests can be coupled with the use ofNEI pressure-washed debris beds in the head loss columns to determine whether the head losscharacteristics in response to direct nucleation of precipitates in the fiber bed are similar to theresponse of precipitates that form after the addition of aluminum nitrate.Document No: CHLE-010, Rev 2Page 31 of 32 Title: CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum AdditionReferences
- 1. Nuclear Energy Institute (NEI). "ZOI Fibrous Debris Preparation:
Processing,
- Storage, andHandling, Revision 1", January 2012.2. Ruland, W.H. Letter to John Butler of the Nuclear Energy Institute with the subject line "FibrousDebris Preparation procedure for Emerengy Core Cooling System Recirculation Sump StrainerTesting, Revision 1" dated April 26, 2012.3. University of New Mexico (UNM). "CHLE-008 Debris Bed Formation
- Results, Rev 3", June 2012.4. IPD Testing Experts.
"Test Report: Report to Timothy Sandy of Alion Science and Technology" dated July 26, 2012.Document No: CHLE-010, Rev 2Page 32 of 32 CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition:
Correlated Control Charts for Head Loss Response to Aluminum Additionby Jeremy Tejada and David Morton, The University of Texas at AustinAugust 17, 20121. Summary of AnalysisWe develop and analyze control charts for the head loss data in Report CHLE-010:
CHLE Tank TestResults for Blended and NEI Fiber Beds with Aluminum Addition (Howe, 2012). Table 1 summarizes the results that we describe in detail in subsequent sections.
The analysis consists of assessing theperformance of control charts for head loss for three columns using, in turn, three CHLE head losscolumns for the NEI fiber beds and three CHLE head loss columns for the blended fiber beds.NEI Fiber BedBlended Fiber BedColumn I Lack of evidence suggesting change in head Strong statistical evidence of change in head lossloss process after addition of aluminum, process after aluminum is added.Statistically significant change in head lossColumn 2 process after addition of aluminum.
Strong statistical evidence of change in head lossConclusion based violations of upper control process after aluminum is added.limit over multiple half-day time periods.Column 3Lack of evidence suggesting change in headloss process after addition of aluminum.
Statistically significant change in head lossprocess after addition of aluminum.
Conclusion based violations of upper control limit using half-day moving average.Table 1. Summary of Results of Analyzing Control Charts for Temperature-corrected Head Loss Processin Three Columns using NEI Fiber Bed and Blended Fiber Bed.2. Introduction The objective of this analysis is to determine whether the addition of aluminum over time caused astatistically significant response in terms of head loss through the debris beds. The experiment wasIAugust 17, 2012I designed to record data each minute, with each data point being the average of six measurements duringthat minute. Thus, the data from the experiment forms a time series where the (average) head loss throughthe debris bed is provided for each minute over several days of experimentation.
The experiment usedthree identical
- physical, vertical
- columns, and data are provided for each column separately.
Thus weperform the analysis we describe on each column separately.
We report analysis for six data sets: ThreeCHLE head loss columns for the NEI fiber beds and three CHLE head loss columns for the blended fiberbeds. In order to assess whether we observe a statistically significant response in head loss due to theaddition of aluminum over time, we use control charts. Specifically, we use control charts designed tohandle correlations and trends in time series data (Croux et al., 2010). We organize the remainder of thisanalysis as follows:
Section 3 presents some of the details of the time series control charts that we use,and Sections 4 and 5 present the results for the NEI fiber beds and the blended fiber beds, respectively, along with a brief discussion of these results.3. Control Charts for Time Series DataIn this section, we present some of the details for a control chart for a time series, {y, }, and ourdescription follows closely that of Croux et al. (2010). At each point in time we obtain a one-step-ahead forecast using the Holt-Winters forecasting algorithm, and we plot the errors from this one-step-ahead forecast on a control chart. We construct the control limits to monitor new observations from a trainingsample, and we use the one day (1440 minutes) before aluminum was added in the experiment as ourtraining period, as we assume the error process has reached a steady state by this point (about 5.8 daysinto the experiment).
3.1. Holt-Winters Forecasting Algorithm Assume we have observed a time series up to time period t -1. The Holt-Winters method predicts theseries at time t, which we denote Y,,- After we observe the actual value, y,, we compute the one-step-ahead forecast error e, according toe,=y,-Y 1We denote the level of the time series by a,, and the trend in the series by /A,. The Holt-Winters methodfor estimating these values after each realization, y,, is as follows2 August 17, 2012I a, = A~y, + (1-A )(a,_,- fl,_) (2)'3 A 2(a t) l (3)and this results in the forecastYO-,--= at-i + f/,-1" (4)The parameters A1 and X2 in (2)-(3) are smoothing parameters that take on values between zero and one.Larger parameters values lead to less smoothing and more weight being placed on the current valuesopposed to the previous level and trend. The relations (2)-(3) begin after a warm up period, in our caseabout 5.8 days. A linear model is fitted to the warm-up data in order to determine estimates of the initiallevel a, and trend /60. After the warm-up period, there is a training period, in our case 1 day or 1440minutes.
We determine the values for the smoothing parameters
), and A. by minimizing the sum ofsquared forecast errors over a training period as follows:(),°P' A"')E- arg min (y (5)1' ,"2 , ý 7, (' -y -'V )" 5A simple two dimensional guess-and-check search over these parameter values can provide the optimalvalues to within two decimal places very quickly.3.2. Control ChartWe monitor and plot the one-step ahead forecast errors on the control chart in order to determine if, andwhen, the process goes out of statistical control.
As with most control charts, we assume the forecasterrors (not the data itself) are normally distributed.
The upper control limit (UCL) and lower control limit(LCL) are set using the forecast errors during the training period. We denote the mean of the squaredprediction errors in the training period bys2= e, (6)nwhere n is the number of data points in the training period. The target value for the forecast errors iszero, so the (1- a) -level UCL and LCL are:3August 17, 2012 UCL = z,/,2 *SLCL = -Zai2 * (7)where zi/2 is the (1-a /2)-level quantile of the standard normal distribution.
For each new observation in the test period, the recursive relations (2)-(3) are updated, and the forecasterror for that observation can be computed and plotted on the control chart. If it falls outside the limits, itsvalue is statistically significantly different from the predicted value, indicating some unexpected changedin the process.
Observations that are part of the test period influence future forecast errors via the Holt-Winters forecast formulas as these are continually updated.
- However, these observations do not alter thecontrol limits.In what follows, we build two-sided control charts as we describe above, using (7). That said, we haveinterest in increases in head loss, and so we pay particular attention to violations of the upper control limitin equation (7). Some violation of that control limit is expected, even if the system remains "in control"due to the statistical nature of the process.
Such violations of the upper control limit would occur at ratea / 2. For this reason, when a more nuanced analysis is needed (Section 4 and Section 5.3), we track amoving average of the rate of violation to help assess whether the head loss process has indeed changedafter the addition of aluminum.
- 4. Control Charts for NEI Fiber Bed Data and Discussion For each of the three CHLE head loss columns with the NEI fiber bed, we present the control chart with a= 0.05, both with the initial training period included and only after aluminum addition.
For the NEI fiberbeds, based on a preliminary
- analysis, it appeared that there may be responses at two points in time afteraluminum was added. Thus, we consider a second training period starting at day 10.5 and ending at day11.5, and we set new control limits for the process after that point. We also give the estimated A, and A2values for both training periods and the observation periods.
- Finally, we discuss the meaning of the chartsfor each column.4.1 Column 1Period 1Training Fraction Outside Control Limits (two-sided)
= 0.0694Training Fraction Outside Upper Control Limit = 0.0417Observed Fraction Outside Control Limits (two-sided)
= 0.09534August 17, 2012 Observed Fraction Outside Upper Control Limit = 0.0126= 0.12A2 = 0.43Period 2Training Fraction Outside Control Limits (two-sided)
= 0.0063Training Fraction Outside Upper Control Limit = 0.0000Observed Fraction Outside Control Limits (two-sided)
= 0.0028Observed Fraction Outside Upper Control Limit = 0.0021A = 0.07A2 =0.10NEI Fiber Temperature-Corrected Head LossControl Chart: Column 10 .0 5 ... .. .. .. ...... ...0.040.03 ... .4 ErrorsS0.02 .... ... CUj- LCL0.01-0.01 -MA(60)-0.02-Alpha-0.03 ... MA(720)TRANING PERIOD 1 TRANING PERIOD 2-0.04 4 ...-0.05 +0 1 2 3 4 5 6 7 8 9 10 11 12 13 14Time (days)Figure 2. NEI Fiber Temperature-Corrected Head Loss Control Chart Column 1 -Initial Training PeriodIncluded.
The figure depicts errors (see equation (1)), the lower control limit and the upper control limit(see equation (7)), a moving average of errors based on the last 60 one-minute
- periods, and a movingaverage of violations of the upper control limit for the last 720 one-minute periods.
Subsequent figureshave similar formatting.
5August 17, 2012 Ww4.,00.050.040.030.020.010.00-0.01-0.02-0.03-0.04-0.05NEI Fiber -Head Loss After Aluminum AdditionControl Chart: Column 1ALSALARS FALSEALARMS
--T -Errorsý -LCLL MA(60)-AlphaMA(720)TRANING PERIOD 2 7}6 7 8 9 10 11rime(days) 12 13 14Figure 3. NEI Fiber Temperature-Corrected Head Loss Control Chart Column 1 -Initial Training PeriodNot IncludedDuring the initial training period, the observed alpha value (i.e., the fraction of observations outside thetwo-sided control limits) of 0.0694 is slightly higher than the expected 0.05, but within reason, leading usto conclude the control limits are reasonable.
During the initial observation period, the observed alphavalue for two-sided violation rose to 0.0943. However, the fraction of violations of the upper control limitare only 0.0126, well within the expected value of 0.025. The only exception to this for a narrower timewindow is the denser mass of errors near the upper control limit depicted early in day 8; see Figure 2. Themoving average of the rate of the upper control limit violation over half-a-day (720 minutes) grows toabout 0.04, compared to an expected upper limit violation of 0.025, but we regard this as being at most aweak signal. An analogous analysis for the second period gives no indication of a signal. Based on this,we do not find compelling evidence that in column 1 there was a statistically significant increase in headloss as aluminum was added over time.4.2 Column 2Period 1Training Fraction Outside Control Limits (two-sided)
= 0.09586August 17, 2012I Training Fraction Outside Upper Control Limit = 0.0243Observed Fraction Outside Control Limits (two-sided)
= 0.0892Observed Fraction Outside Upper Control Limit = 0.0451S= 0.23A2= 0.00Period 2Training Fraction Outside Control Limits (two-sided)
= 0.0903Training Fraction Outside Upper Control Limit = 0.0417Observed Fraction Outside Control Limits (two-sided)
= 0.1091Observed Fraction Outside Upper Control Limit = 0.0539,A =0.15A2=-0.01NEI Fiber Temperature-Corrected Head LossControl Chart: Column 20.05 .0.040.03 ....2 0 .0 2 -... .. ... ..... ... ..'A1j i.0.01S-0.01 0-0.020.0TRANING PERIOD 1-0.04 ......-0.05 + ...... ....0 1 2 3 4 5TRANING PERIOD 2-Errors-LCLI UCL.MA(60)-AlphaMA(720)13 147 8 9 10 11Time (days)12Figure 4. NEI Fiber Temperature-Corrected Head Loss Control Chart Column 2 -Initial Training PeriodIncluded7August 17, 2012 NEI Fiber -Head Loss After Aluminum AdditionControl Chart: Column 20.056 LS SIGNALS0.040.03 ---Errors0.020.01 -LCL0.00o0.00 --UCL-0.01--0. MA(60)-0.02-0.03 -Alpha-0.04 TRANING PERIOD 2 MA(720)-0.05...6 7 8 9 10 11 12 13 14Time(days)
Figure 5. NEI Fiber Temperature-Corrected Head Loss Control Chart Column 2 -Initial Training PeriodNot IncludedDuring the first training period, the observed alpha value of 0.0958 is again a bit higher than the expected0.05, but perhaps close to within reason. The observed upper control limit violations of 0.0243 matchesclosely the expected 0.025 during the first training period. During the first observation period, theobserved alpha value for upper control limit violations climbs to a modest value of 0.0451. However, asthe moving-average values in Figure 4 depict, for substantial portions of day 8 and days 9-10.4, there aresignificant violations of the upper control limit. The associated "blue line" exceeds the chart's maximum,and peaks at 0.0625 during the first such interval (day 8) and 0.0722 during the second interval.
There issimilar behavior during the second period. So while the process is not consistently out of statistical control after the addition of aluminum, there are significant periods of more than one-half day in lengthwhere the process deviates well beyond the control limits. This suggests a statistically significant deviation of the process after the addition of aluminum.
We further note that the timing of thesedeviations is consistent with visible jumps in the level of the head loss for column 2 in Figure 2 of Howe(2012).4.3 Column 38August 17, 2012I Period 1Training Fraction Outside Control Limits (two-sided)
= 0.0271Training Fraction Outside Upper Control Limit = 0.0146Observed Fraction Outside Control Limits (two-sided)
= 0.0202Observed Fraction Outside Upper Control Limit = 0.0130/3 =0.11= 0.32Period 2Training Fraction Outside Control Limits (two-sided)
= 0.0125Training Fraction Outside Upper Control Limit = 0.0097Observed Fraction Outside Control Limits (two-sided)
= 0.0135Observed Fraction Outside Upper Control Limit = 0.0078.A =0.102 =0.13NEI Fiber Temperature-Corrected Head LossControl Chart: Column 30.050.040.03 Errors-0.03 -- MA(6-0)-.4i lTRANING PERIOD 1 TRANING PERIOD 2 ....--0.04-0.05 1 .0 1 2 3 4 5 6 7 8 9 10 11 12 13 14Time (days)Figure 6. NEI Fiber Temperature-Corrected Head Loss Control Chart Column 3 -Initial Training PeriodIncluded9August 17, 2012 NEI Fiber -Head Loss After Aluminum AdditionControl Chart: Column 30.05 s0.040.03 -Errors0.02-S-- LCL0.010.00 UCL-0.01 -MA(60)0- -0.02L --Alpha-0.03 .MA(720)-0.04 TRANING PERIOD 2-0.056 7 8 9 10 11 12 13 14Time(days)
Figure 7. NEI Fiber Temperature-Corrected Head Loss Control Chart Column 3 -Initial Training PeriodNot IncludedDuring both training periods and observation
- periods, the observed alpha values are less than 0.05 forviolations of the two-sided control limits and less than 0.025 for violations of the upper control limit.Individual errors violate control limits only on rare occasions.
The moving-average of violations of theupper control limit modestly exceeds 0.025 for part of day 7,'but otherwise is almost exclusively below0.025. It is reasonable to conclude that in column 3, there was not a statistically significant increase inhead loss as aluminum was added over time.5. Control Charts for Blended Fiber Bed Data and Discussion Again for each of the three columns under the blended fiber bed data, we present control charts with a =0.05 both with the initial training period included and only after the addition of aluminum.
For theblended fiber beds, we only use one training period and one observation period. We again provide theobserved alpha values and optimal smoothing parameters for both the training and observation period.The results here are not subtle for columns 1 and 2 and a formal statistical analysis is arguably overkill inthese cases. (See Figure 1 of Howe, 2012.) We simplify the discussion slightly by not distinguishing two-10August 17, 2012I sided and upper-limit violations, and by suppressing presentation of the upper control limit violation, atleast until the discussion in Section 5.3.5.1 Column 1Training Period Fraction Outside Control Limits = 0.0454Observation Period Fraction Outside Control Limits = 0.3832A1 =0.16A2 =0.48Blended Fiber Temperature-corrected Head LossControl Chart: Column 100.1 -ErrorsLU0.05 .... ..----- LCL2-UCL-0.05 MA(60)" -0 .1 .-0 .1 5 4--. .. ..+. ...... .. .. ...... ................ ....5.5 5.75 6 6.25 6.5 6.75 7 7.25 7.5 7.75 8 8.25 8.5lime (day)Figure 8. Blended Fiber Temperature-Corrected Head Loss Control Chart Column 1 -Initial TrainingPeriod IncludedI1IAugust 17, 2012 Blended Fiber Head Loss After Aluminum AdditionControl Chart: Column 10.151A2 0.1 -__- Errors0.05o0 --LCL.Q 0 0S -0 .0 5 i ..--- U C L-0.1 4.-0.15 4- .. .-.6.6 6.8 7 7.2 Timq.Way) 7.6 7.8 8 8.2Figure 9. Blended Fiber Temperature-Corrected Head Loss Control Chart Column 1 -Initial TrainingPeriod Not IncludedThe observed alpha value during the training period of 0.045 is close to the expected 0.05, leading us toconclude the control limits set during training are sound. Examining Figure 7, the steady state assumption of the error process during the training period appears appropriate.
From both Figures 7 and 8, we observean almost immediate response after aluminum is added, with a delay of about 0.05 days. The one-hourmoving average captures this change fairly well, and the errors quickly exceed the upper control limit. Atabout 7.3 days, the errors increase significantly and the process jumps to a further out-of-control state. Incolumn 1, there is clearly a statistically significant response to the addition of aluminum, much morepronounced than with any of the NEI fiber beds.5.2 Column 2Training Period Fraction Outside Control Limits = 0.0477Observation Period Fraction Outside Control Limits = 0.2462A =-0.40, =0.0012August 17, 2012I Blended Fiber Temperature-corrected Head LossControl Chart: Column 20.58 0.3 .......- Errors.- 04 , ..0.1 -LCL--UCLU 0.1 --0.2 .MA(60)a, -0.23-0.4 .5.5 5.75 6 6.25 6.5 6.75 7 7.25 7.5 7.75 8 8.25 8.5Time (day)Figure 10. Blended Fiber Temperature-Corrected Head Loss Control Chart Column 2 -Initial TrainingPeriod IncludedBlended Fiber Head Loss After Aluminum AdditionControl Chart: Column 20.51 0.40 0.3!M 0. Errors0.20.1 LC/-UCL-0 .1 i ..... ... .-0.2 ........
MA(60)-0.3-0.4-0.5 4-6.6 6.8 7 7.2 7.4 7.6 7.8 8 8.2Time (day)Figure 11. Blended Fiber Temperature-Corrected Head Loss Control Chart Column 2 -Initial TrainingPeriod Not IncludedThe observed forecast errors for column 2 are very similar to those of column 1. The observed alphavalue during the training period is 0.045. From Figure 9, we see that the error process during the training13August 17, 2012 period is relatively stable, consistent with a steady state error process.
Examining Figure 10, we seeanother immediate response after aluminum is added, with a delay of about 0.05 days. The one-hourmoving average again captures this change fairly well and the errors quickly exceed the upper controllimit. For this column, the error process appears to stabilize over the next day, and then at day 7.7 anotherlarge shift above the upper control limit occurs. In column 2, there is clearly a statistically significant response to the addition of aluminum.
5.3 Column 3Training Period Fraction Outside Control Limits = 0.0293Observation Period Fraction Outside Control Limits = 0.0525--0.60A2 =0.19Blended Fiber Temperature-corrected Head LossControl Chart: Column 30.75I 0.5 -Errors0-UCLS-0.25 ....-MA(60)0S-0.5-0.75 .5.5 5.75 6 6.25 6.5 6.75 7 7.25 7.5 7.75 8 8.25 8.5Time (day)Figure 12. Blended Fiber Temperature-Corrected Head Loss Control Chart Column 3 -Initial TrainingPeriod Included14August 17, 2012 IBlended Fiber Head Loss After Aluminum AdditionControl Chart: Column 30.75 -0 0.5- --Uj Errors0.25 .-LCL0-UCL-0.25 .MA(60)U6 -0.54-0.75 -6.6 6.8 7 7.2 7.4 7.6 7.8 8 8.2lime (day)Figure 13. Blended Fiber Temperature-Corrected Head Loss Control Chart Column 3 -Initial TrainingPeriod Not IncludedExamining Figure I in Howe (2012) for column 3, we see an apparent change in the process, i.e., anapparent change in slope. However, compared to columns 1 and 2 for the blended fiber beds, the volatility of column 3's process makes this more difficult to establish statistically.
In contrast to columns 1 and 2,the observed violation of the two-sided control limits for column 3 is 0.0525. The fraction of violation ofthe upper control limit is 0.0324. However, when forming a moving average of the upper control limitviolation of 720 minutes (not depicted in the figures) we find that the violation steadily climbs fromaround 0.014 at day 6.33 to 0.064 at day 7.25; i.e., the moving average climbs to a value substantially larger than the expected value of 0.025. While more subtle than columns 1 and 2, this suggests a changein the head loss process after the addition of aluminum.
That said, the change is only statistically significant-according to the moving average measure-substantially later than for columns 1 and 2.References Croux, C., Gelper, S., & Mahieu, K. (2010 1-April).
Robust Control Charts for Time Series Data.Retrieved 2012 5-August from http://dx.doi.org/l0.2139/ssm.1588646 Howe, K. (2012). CHLE Tank Test Results for Blended and NEI Fiber Beds with Aluminum Addition.
University of New Mexico, Department of Civil Engineering, Albuquerque.
15August 17, 2012