ML20211F661
| ML20211F661 | |
| Person / Time | |
|---|---|
| Issue date: | 05/19/1986 |
| From: | Telford J NRC OFFICE OF NUCLEAR REGULATORY RESEARCH (RES) |
| To: | Curtis R NRC OFFICE OF NUCLEAR REGULATORY RESEARCH (RES) |
| Shared Package | |
| ML20211F389 | List:
|
| References | |
| FOIA-87-8 NUDOCS 8702250163 | |
| Download: ML20211F661 (12) | |
Text
~
l y,
t
~ Jl -
e E 19 M MEMORANDUM FOR:
Robert T. Curtis, Chief Containment Systems Research Branch FROM:
John L. Telford Containment Systems Research Branch
SUBJECT:
SURTSEY TEST MATRIX FOR PRESSURIZED MELT EJECTION AND DIRECT CONTAINMENT HEATING In a letter from Mr. Robert B. Minogue to Mr. A. W. Snyder of SNL, dated November 16, 1984, a request was made "to incorporate statistically designed experiments and other statistical techniques as a routine way of doing business in NRC-sponsored research." As part of the effort to continue using designed experiments to obtain maximum inforTnation for dollars spent, and in response to a request for comments on the DCH research programs from you, I have reviewed the FY 86-87 Program Plan for DCH [1] which was discussed on April 22 & 23, 1986 at the two DCH meetings.
The bottom line is there are several better alternative designs than the one proposed in the program plan. The basic problems with the proposed test matrix (copy enclosed) are that too many experimental variables are investigated for too few tests and there are no replicate tests to quantify experimental variability in the results.
I feel that valid inferences based on the proposed tests will be difficult if not impossible.
In the following I will review:
the meeting advice, the needs of the experimental program, the problems with the proposed test matrix, and some better options.
Meetings During the two DCH meetings the following advice was given by several people.
1.
For some reactors the cavity design will not be an effective mitigative factor for DCH.
l 2.
The effect of structure in three dimensions in the containment above the lower cavity may mitigate DCH.
3.
The effect of water in the atmosphere inside the containment may mitigate DCH.
We can interpret that these three effects: cavity design, 3D structure in containment, and water in the atmosphere are independent variables that should be in the final test matrix.
l g 22 g 3 070217 GHOLLyB7-8 pg l
I
\\
- y 8bbertT.Curtis 2
'h
\\
Experimental Program Needs p.
34-As discussed in the FY 86-87 DCH Program Plan (p. 28) and the SNL meeting Y
handouts (pp. 24-26) the needs are the following.
e L
/
1.
We need to be able to interpret (analyze) the effects (phenomena) observed i
in the experiments.
Part of the interpretation is being able to declare that an effect is "large" (or significant) in the presence of experimental variation. This ability can easily be built into an efficient
.Q experimental matrix.
4 2.
We need to be able to develop (fit) physical models by estimating the coefficients in the model proposed by Dr. M. Pilch.
The model depends on one mass' transfer coefficient, four heat transfer coefficients, and the material properties assuming a Kutateladze number greater than 20 causes entrainment. This may be the most important need, given that all relevant combinations:of effects (phenomena) are present in the test matrix.
The overall need is, for minimum cost, to reach an understanding of the phenomena sufficient to solve the DCH issue for reactor safety. A carefully designed effici.ent test matrix is required.
Problems With Proposed Test Matrix If we examine the eleven proposed tests, to ask if the measured results can be
- /
'used to estimate one of the needed model coefficients, the following table wcpid emerge.
1 TABLE 1 i
Coefficients to Estimate (h) 5 Debris to Debris to Atmosphere Atmosphere Test Mass Atmosphere Water to Water to Structure 1
M1 No No No No 2
M2 DAl No No No 3
M3 DAl No No No 4.
M4 DA1 No No ASl Si MS DA1 No No AS2 6
M2 DA2 No No No 7
M2 DA3 DW1 AW1 No 8
M2 DA4 DW2 AW2 No 9
M6 DAS No No No 10 M7 DA4 DW3 AW3 No 11 M8 DA4 DW4 AW4 No i
l l
c
Robert T. Curtis 3
where:
Mi = the mass transfer coefficient estimate for the i,t_h test.
h No = not used or of questionable applicability DAi = the debris to atmosphere heat transfer coefficient estimate for the i M test.
DWi = the debris to water heat transfer coefficient estimate for the iM test.
AWi = the atmosphere to water heat transfer coefficient estimate for the i M test.
ASi = the atmosphere to structure heat transfer coefficient estimate for the i M test.
The advantages, disadvantages, and questions are:
1.
There could be eight different mass transfer coefficient estimates due to the effect of: mass, cavity geometry, containment structures, melt material, and water in the cavity. What is the overall estimate and its variability?
2.
Tests 6, 7, and 8 may be repeats of test 2, an advantage in estimating experimental variability for mass transfer.
3.
There could be five different debris to atmosphere heat transfer coefficient estimates due to the effect of:
containment atmosphere gas, water in the atmosphere, and melt material. What is the overall estimate and its variability?
4.
Tests 3, 4, and 5 may be repeats of test 2, an advantage in estimating experimental variability for debris to atmosphere heat transfer. Tests 10 and 11 may be repeats of test 8 for debris to atmosphere heat transfer.
5.
Test 7 has a possible confounding effect of an additional source of heat from hydrogen. Are there any comparison tests with air and steam but no hydrogen?
6.
As far as debris to water heat transfer is concerned the first six tests and test 9 are of n_o value (no water). This provides no information for seven tests.
7.
There could be four different debris to water heat transfer coefficient estimates due to the effect of:
steam and hydrogen, water sprays, and amount of water in the cavity. What is the overall estimate and its variability?
Robert T. Curtis 4
8.
Tests one to six and nine are of no value as far as atmosphere to water j
heat transfer is concerned (no waIer). This provides no information for seven tests.
9.
There could be four different atmosphere to water heat transfer coefficient estimates due to the effect of: water sprays, steam and hydrogen, and amount of water in the cavity. What is the overall estimate and its variability?
10.
The coefficients for debris to water and atmosphere to water are confounded and correlated since any changes in the experimental variables are made simultaneously for both heat transfer coefficents. This is a major mistake.
- 11. Tests one to three and six to eleven are of no value as far as atmosphere to structure heat transfer is concerned (no structure). This provides no information for nine tests.
- 12. What are we assuming for debris to structure heat transfer?
- 13. Note that the variability in each of the five transfer coefficient estimates will be induced by a different set of sources. This is an unwelcomed problem.
In summary, the proposed test matrix is based on the Zion cavity, 80 kg melt, and 4 MPa pressure. The possible comparisons for analysis of measured results are based on "a base case" of test 2 and the logic of changing one-factor-at-a-time (with the exception of test 7). This logic yields pair wise comparisons of test results and very inefficient use of test dollars as noted in items 6, 8, and 11 above.
The proposed test matrix devotes some effort to estimating the effect of: mass scale, cavity, structures, inert gas atmosphere, air and steam and hydrogen atmosphere, air atmosphere and water sprays, melt material, and water in the cavity.
The worst part is all of the relevant experimental variable combinations are not in the proposed test matrix, as we shall see.
If we are interested in the effects of structure and water on the heat transfer coefficients for debris-water, atmosphere-water, and atmosphere-structure, as the DCH meeting participants seemed to be, then let's examine the proposed experimental variable combinations.
For two experimental variables of interest, structure and water, the following table provides the variable combinations in the proposed test matrix.
TABLE 2 Proposed Variable of Interest Test Number Structure Water None yes yes 8 and 10 no yes 4 and 5 yes no 2 and 6 no no
i Robert T. Curtis 5
This table also provides the required combinations to investigate these two experimental variables. Thus we will be able to estimate the effect of either structure or water, but not be able to estimate the interaction effect of structure and water.
We should consider any possible way in which water and structure could lead to increased mitigation, e.g., the effect of sprays wetting the structure causing an increased heat transfer to the water. This design flaw in the proposed test matrix could easily be fixed by including both structure and water in one of the actual tests.
If we are interested in the effects of cavity and structure, on the transfer coefficients for mass, debris-atmosphere, and atmosphere-structure, as the DCH meeting participants seemed to be, then let's examine the proposed experimental variable combinations.
For two experimental variables of interest, cavity and structure, the following table provides the variable combinations in the proposed test matrix.
TABLE 3 Proposed Variable of Interest Test Number Cavity Structure 4 and 5 Zion Yes None Surry yes 2
Zion no 3
Surry no This table also provides the required combinations to investigate these two experimental variables. Thus we will be able to estimate the effect of either cavity or structure, but not be able to estimate the interaction effect of cavity and structure. We should consider any possible way in which cavity and structure could lead to increased mitigation, e.g., the effect of a Surry cavity and structure to reduce the velocity and mass transfer. This design flaw in the proposed test matrix could easily be fixed by including both cavity and structure in one of the actual tests.
If we are interested in the effects of cavity, structure, and water on the mass transfer and four heat transfer coefficients, than let's examine the proposed experimental variable combinations. The following table provides the variable combinations in the proposed test matrix.
6 Robert T. Curtis 6
TABLE 4 Proposed Varialbe of Interest Test Number Cavity Structure Water CSW None Zion yes yes
+
None Surry yes yes 8 and 10 Zion no yes None Surry no yes
+
4 and 5 Zion yes no None Surry yes no
+
2 and 6 Zion no no
+
3 Surry no no This table also provides the required combinations to investigate (quantify) the effects of:
cavity, structure, water, and the interactions of:
cavity-structure, cavity-water, and structure-water on the mass transfer and four heat transfer coefficients.
This table also provides the one-half fractions (4 tests) of the full factorial matrix (8 tests). Thus, we will not be able to estimate the effects of cavity, structure, and water (i.e., the main effects) after doing the 11 proposed tests. This is because a necessary one-half fraction test design was not selected as can be seen from the column labeled "CSW".
We need either all combinations (4) with a "+" in that row or all combinations (4) with a " " in that row, for a one-half fraction design.
This is a serious design flaw.
Further, we will not be able to estimate the interaction effects of:
cavity-structure, cavity-water and structure-water.
We should not accept this when we strongly suspect (know?) that two of the interactions may be important.
On page 26 of the FY 86-87 DCH Program Plan it states, "A statistically based testing approach will insure that the effect of these initial conditions over their range is evaluated.
The product of such a procu s would describe both the direct influence and the interactions of the initial conditions. The listing in Table 5 shows that there are fourteen initial conditions to be considered.
A simple two-factorial test matrix would require over 10,000 experiments... A method is necessary for reducing the number of variables being considered."
My responses are as follows.
1.
The proposed matrix is not "a statistically based testing approach" and suffers from several flaws as a result (as described above in itens 1, 3, 5, 6, 7, 8, 9, 10, 11, and 13).
2
i Robert T. Curtis 7
4 2.
Table 5 does contain 14 " initial conditions" but the proposed test matrix selects one constant value for: gas pressure, fraction of core melted, melt temperature, fraction of metals oxidized, dissolved gas, annular RPV gap, in-cavity structures, and vessel pressure. This leaves six experimental variables.
3.
A test matrix to investigate six experimental variables [2] requires six
" pieces of information" (tests) for main effects,15 " pieces of information" for two way interactions, one " piece of information" for the overall mean, for a minimum of 22 (total number of) tests. The closest fractional factorial " test matrix" design is 32 tests, a one-half fraction for six variables each at two labels (unless we get clever and lucky using a smaller fraction of 16 tests plus 8 tests to unconfound the necessary two way interactions).
4.
A test matrix for 14 experimental variables [2] requires 128 tests:
14 for main effects, 91 for two way interactions, one for the overall mean, and 22 for estimating experimental variability. The proposal (p. 26) should not claim that 10,000 experiments are required.
5.
We do need to reduce the number of experimental variables. We should reduce the 14 variables to no more than five variables. A test matrix for five experimental variables requires 16 tests. Since 11 tests are straining the budget, it would be an advantage to reduce to four variables. We reduce to four (or five) variables by rank ordering the candidate variables in terms of importance according to what we know about the physics and what we have learned over the last couple of years of testing.
Conclusions 1.
We should try to estimate the transfer coefficients using the Pilch model.
2.
We should not use the proposed test matrix or any other change-one-variable-at-a-time approach, because it is grossly inefficient and decisions will be made based on a comparison of the base case (e.g.,
test 2) and one variable change (e.g., test 4 or test 8). How do we know the observed response is important and not due to chance or experimental variation? After all, test "A" versus test "B" is a sample of size one for each test. There are no replicate tests in the proposed test matrix to quantify experimental variability.
Robert T. Curtis 8
3.
We need to reduce the number T.? experimental variables to no more than four in order to have defensible conclusions for the allowed cost. The DCH Review Group should recomend the four (or three) most important variables to be used in the tests.
4.
We need to carefully examine the. inferences that can be made based on the aerosol portion of these tests. Most of the tests have a dry vessel atmosphere. We know that very high relative humidity levels cause chain-like aerosol agglomerates to change the sphere-like aerosol agglomerates. So of what value are the dry tests? The Zion containment will not be dry. The method described for estimating the aerosol shape factors would seem to yield non-unique answers (p.18), especially since this is a difficult task under the best conditions.
5.
We should consider including in cavity structure, e.g., steel braces, pipes, ladders, and grids, as part of our containment " structure." The real reactor cavities are very cramped and not clean of structure at all.
After all we are trying to build a model to use for the reactor case.
Recommendations 1.
If we want three experimental variables (e.g., cavity, structure, and water, tTie following test matrix should be considered.
TABLE 5 Experimental Variable Reference to SNL Test Cavity Structure Water Proposed Test Number 1.
Zion yes yes None 2.
Surry yes yes None 3.
Zion no yes 8 and 10 4.
Surry no yes None 5.
Zion yes no 4 and 5 6.
Surry yes no None 7.
Zion no no 2 and 6 8.
Surry no no 3
9.
Use corium melt in the same experimental conditions as test 1 or 2.
- 10. Use a 20 kg melt mass in the same experimental conditions as test 1 or 2.
Robert T. Curtis 9
The test run order should be random. The DCH Review Group should re-commend what three dimensional structure is put in the test matrix where a "yes" appears in the structure column and where and how much water is used in the test matrix where a "yes" appears in the water column.
This test maxtrix design allows the use of all eight test results to estimate the effects of cavity, structure, water, and all two-way interactions.
Compare this to one test result versus one test result.
This advantage is called hidden replication.
This test matrix saves the cost of one test (10 not 11) and allows the efficient estimation of all mass and heat transfer coefficients (refer to Table 1 above).
i 2.
If we want four experimental variables as on page 26 of the SNL Plan Te.g., cavity, structure, water, and mass), the following test matrix should be considered.
TABLE 6 Experimental Variables Test Cavity Structure Water Mass CSWM 1
Zion yes yes 80
+
2 Surry yes yes 80 3
Zion no yes 80 4
Surry no yes 80
+
5 Zion yes no 80 6
Surry yes no 80
+
7 Zion no no 80
+
8 Surry no no 80 9
Zion yes yes 20 10 Surry yes yes 20
+
11 Zion no yes 20
+
12 Surry no yes 20 13 Zion yes no 20
+
14 Surry yes no 20 15 Zion no no 20 16 Surry no no 20
+
Robert T. Curtis 10 P
The test matrix would be either the "+" or " " half as defined in the CSWM 7 column. The test run order should be random.
This set of eight tests would allow estimation of the four main effects but not the six two-way interactions. This is the penalty of using four variables.
As before, the DCH Review Group should recomend how the variable levels are defined (e.g., "yes" for structure, etc.). An additional 2 or 3 tests could be added to these eight tests, as required, e.g., using corium.
3.
Above all, don't use a change-one-variable-at-a-time approach.
Use an efficient design with hidden replication to estimate the effect of the experimental variability and to estimate the mass and heat transfer coefficients for the Pilch model.
4.
The proposed test matrix attemps to investigate too many experimental variables (e.g., it seems like more than six) in too few tests with no replicate tests to quantify experimental variability.
We should not attempt to investigate more than four experimental variables for our current budget which allows 11 tests.
In order to do an adequate job with more than four experimental variables, we would need more tests.
5.
We need to be satisfied, before testing starts, that the number of experimental variables, assumptions, cost, and possible post-testing inferences and decisions are acceptable. We can accomplish this through careful review and test matrix design.
The draft FY 86-87 Program Plan for DCH has served its purpose. The authors deserve our congratulations for a good first effort, especially the proposed model.
Since the first story teller doesn't have a chance to survive, from here improvements should be easy.
John L. Telford Containment Systems Research Branch
Enclosure:
As stated cc:
R. Sehgal, EPRI DISTRIBUTION: circ; chron; Curtis; Telford; A. W. Snyder, SNL Morris; Conti; Ross; Minogue; W. Tarbell, SNL Speis; Lee; Silberberg; Burson; M. Pilch, SNL Rasmuson; Wright; Chan; J. Brockman, SNL Worthington; Wood; Eltawila D. Powers, SNL, panel member M. Corradini, U. of WI, panel member B. Sper.cer, ANL, panel member T. Ginsberg, BNL, panel member T. Theofanous, U. of CA, panel member CSRB CSRB TELF0N/lm CURTIS 5/jj/86 Sq/86
Robert T. Curtis 11 REFERENCES 1.
Pressurized Melt Ejection and Direct Containment Heating, HIPS Program Plan for FY 86-87, by William W. Tarbell, Marty Pilch, and John E. Brockman.
2.
Statistics for Experimenters by George E. P. Box, William G. Hunter, and J. Stuart Hunter, John Wiley and Sons, 1978, p. 410 and related.
d
ENCLOSURE 1
~
ROUGH DRAFT Table 6 SURTSEY Test Matrix Test Cavity Mass Pressure Objective Number Model (kg)
(MPa)
Mass Scaling 1
Zion 20 4
Assess mass scale Cavity and Containment Geometry 2
Zion 80 4
Open cavity geometry 3
Surry 80 4
Restricted cavity geometry 4
Zion 80 4
Scaled in-containment structures 5
Zion 80 4
Containment structures of defined size and flow area Energy Exchange 6
Zion 80 4
Inert gas in containment
,f 7
Zion 80 4
Air, steam and hydrogen (2%)
8 Zion 80 4
Air with water sprays 9
Zio'n 80 4
Air atmosphere, corium melt composition Influence of Water 10 Zion 80 4
Water-filled cavity 11 Zion 80 4
Shallow water pool i
l Avi]GH : DRAFT
.. v
.