University
stringclasses 19
values | Text
stringlengths 458
20.7k
|
|---|---|
Colorado School of Mines
|
4.3 Magnetic Separation
TREE grade and recovery data were plotted for the six Wet High Intensity
Magnetic Separation (WHIMS) tests that were conducted. In Figure 4.14 it can be seen
that the TREE grade increased with finer grind size and higher field strength. The figure
also shows that TREE recovery is greatest with the coarsest feed and the highest field
strength. Such a result is expected: as the bastnaesite is magnetized in the separator,
anything that is associated with it will report to the magnetic fraction. When very little is
attached (because the grind size is fine), the magnetic product TREE grade is high –
but when the grind size is coarse, the associated particles also report to that fraction –
and so the coarse grade is low. A similar phenomenon would cause the high recovery in
the coarse product – fine bastnaesite grains were able to pass through the separator
attached to relatively large nonmagnetic grains.
The distribution of the barium between the nonmagnetic and the magnetic
fractions is noteworthy. Figure 4.15 displays that the barium content in the nonmagnetic
fraction is much higher than that in the magnetic fraction. Although the recoveries to the
magnetic product only ranged from 3-17% (see Table 4.6), the potential exists for
further optimization of this process to result in substantial rejection of barite from the
feed.
Table 4.6. Calcium, barium, and TREE grade and recovery of WHIMS concentrate
products.
Field Strength, Grade, Percent Recovery, Percent
Run P , µm Gauss Ca Ba TREE Ca Ba TREE Mass
80
1 762 5000 10 4 2.4 7.1 5.4 10.3 8.8
2 50 5000 8 2 4.9 1.5 0.6 2.9 2.8
3 762 10000 11 3 3.7 17.0 9.3 28.5 16.8
4 144 7500 11 2 3.8 2.6 0.8 4.1 4.0
5 50 10000 14 2 8.4 5.9 1.6 11.5 6.6
6 144 7500 12 3 9.6 3.3 1.3 7.4 4.6
61
|
Colorado School of Mines
|
Figure 4.16. Calcium recoveries to HLA products.
These results were as expected. The concentration criteria (CC from Table 2.4)
for bastnaesite and barite are 2.35 and 2.06. These represent relatively good
separation. Unsurprisingly, these are the minerals that reported to the sink product.
Calcite has a criterion of 1.00, which dictates that a separation in water is fundamentally
impossible. As the fluid became denser, the density difference pushed more of the
calcite to the floats.
For a non-selective process, the mass of each mineral would match that of the
overall mass recovered. If the mineral reports to a product at a greater percentage than
the product represents from the feed, it can be deduced that a preferential separation
exists. Such is the case with barium and the rare earths. The increased liberation with
grind time and low density of calcite, the main calcium-containing mineral in this ore,
explain the increased recovery to the floats at high fluid densities.
This trend continued in the concentrate streams on the Wilfley shaking table. The
reprocessed ore showed good separation into different streams (Figure 4.18). The
compositions of spots 1, 2, 3, and 4 are given in
Table 4.7 below. These concentrations show the motion of the table is successful
in separating the heavier bastnaesite and barite from the lighter calcite. Unfortunately,
damage to the table on the discharge end destroyed the fluid film and prevented
effective separation of a concentrate from a middling product. As the minerals flowed
63
|
Colorado School of Mines
|
the calcium was rejected from the reprocessed concentrate (as only 3% was
recovered).
Figure 4.18. Shaking Table Streams
Table 4.8. Shaking table mass balance for two similar 500 g tests, 2 and 3.
Feed (Calculated) Concentrate
Mass Grade (%) Mass Grade (%) Recovery (%)
Test (g) Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE
2 500 13 14 4.4 53 12 16.18 7.8 11 9 12 19
3 500 13 16 5.1 63 9.3 20 9.2 13 9 16 23
Tailings
Mass Grade (%) Recovery (%)
(g) Ca Ba TREE Mass Ca Ba TREE
2 All recoveries correspond 180 13.9 13.8 3.4 36 37 35 28
3 to a 500 g feed 160 13.8 14 3.4 32 35 29 22
This suggests the possibility of a roughing – cleaning – scavenging circuit to
enrich the concentrate while retrieving as much rare earth content as possible. That
proposed circuit was tested by re-tabling the initial concentrate, middlings, and tails for a
total of eight products (no mass reported to the tails product after reprocessing the initial
65
|
Colorado School of Mines
|
concentrate). The weight percentages and total recoveries with respect to the initial feed
of each of the eight products of the multi-table circuit can be found in APPENDIX C:
EXPERIMENTAL DATA. That table also includes extended versions of Table 4.8 and
Table 4.9, including the middling product and mass balances.
From the 9.2 wt% TREE concentrate (of Test 3), a rare-earth-enriched
concentrate (20%) was created (Test 6). Of the rare earths in the initial feed, 25%
reported to the second concentrate. As only 2% of the calcium gangue reported to the
cleaned concentrate, that product has the best rare earth recovery and lowest calcium
content.
The middling from Test 3 held 38% of the rare earths and produced (through
Test 5) a concentrate, middling, and tails with elevated calcium contents (17%, 16%,
and 15%) and low rare earth contents (2.5%, 3.4%, and 3.5%). The significant mass of
the middling products means that those products still contained 3%, 4%, and 21% of the
rare earths.
Slimes are likely responsible for carrying 5% of the rare earth content into the
tails of the Test 3 tails (during test 4). Only 7% of the REEs reported to that concentrate,
and 19% ended up in its middlings. The concentrate had slightly enriched rare earth
content, at 5.4%, but the middlings and tails only had grades of 1.9% and 3.2% and
high calcium contents (13%, 17%, and 15%).
Based on the grades and recoveries of the different products, a flowsheet (Figure
4.19) might contain roughing tables to reject slimes, cleaning tables to produce a final
concentrate, and scavenging tables to upgrade the middling product. Slimes production
could be controlled in the comminution circuit by raising the grind size, but that would
reduce the liberation factor of the ore. A rod mill (which limits creation of fines in
grinding) and cyclones (for particle classification, to reduce overgrinding of fine
particles) are already employed (Figure 1.1).
66
|
Colorado School of Mines
|
Table 4.9. Mass balance for thrice-tabled concentrate. Products are the final
concentrate and combined total middlings and tailings.
Feed (Calculated) Concentrate
Mass Grade (%) Mass Grade (%) Recovery (%)
(g) Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE
500 13 15 5.3 26 8 21 14 5 3 7 13
Middlings
Recoveries correspond to 500 g feed Mass Grade (%) Recovery (%)
(g) Ca Ba TREE Mass Ca Ba TREE
314 15.5 9 3 63 67 51 50
Tailings
Balance (%) Mass Grade (%) Recovery (%)
Mass Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE
93.32 93.32 93.32 93.32 127 13.5 14 3.6 25 27 23 17
Figure 4.19. Gravity concentration circuit.
Six responses of the Falcon test campaign were recorded: TREE grade, calcium
grade, barium grade, TREE recovery, calcium recovery, and barium recovery. These
responses were analyzed with Design Expert 9. The factorial design of the experiments
allowed for the analysis of the significant effects on the process. Only two responses
produced a signal-to-noise ratio high enough (that is, greater than 4.0) to consider
navigation of the design space: TREE grade and calcium recovery (Table 4.10).
67
|
Colorado School of Mines
|
However, both models (TREE grade and calcium recovery) showed that all of the input
variables were statistically insignificant, and that the models themselves were
insignificant. Their R2 values were .7757 and .7527, respectively. The Analysis of
Variance (ANOVA) table for the TREE grade model is given in Table 4.11.
Table 4.10. Signal-to-Noise Ratio for Falcon DOE Responses
Response Signal-to-Noise Ratio
TREE Recovery 1.61
TREE Grade 6.65
Ca Recovery 4.15
Ca Grade 2.43
Ba Recovery 2.32
Ba Grade 2.14
Table 4.11. ANOVA Table for the TREE Grade Model.
Sum of Mean F p-value
Source Squares df Square Value Prob > F
Model 6.45 5 1.29 3.46 0.0998
A-G-Force 0.18 1 0.18 0.48 0.5181
B-Feed Rate 1.13 1 1.13 3.02 0.1429
C-Feed Size 2 1 2 5.36 0.0684
AB 0.72 1 0.72 1.93 0.2233
BC 2.42 1 2.42 6.49 0.0514
Residual 1.86 5 0.37
Lack of Fit 1.42 3 0.47 2.12 0.3369
Pure Error 0.45 2 0.22
Cor Total 8.31 10 R2 = 0.7757
The grade and recovery from each test is given in Table 4.12. Grade is most
strongly a function of feed size and feed rate, with G-Force having a smaller effect. The
best results came from the low G-Force, low feed rate, and small feed size tests. Tests
68
|
Colorado School of Mines
|
were repeated on those optimum conditions, as they were the ones predicted by the
design model to give the best grade, and those results are given in Table 4.12.
Table 4.12. Grades, recoveries, and concentration ratios of Falcon products.
DOE Falcon G-Force Grind Time Feed Rate Recovery TREE Grade Concentration
Standard Run G's minutes kg/hr Percent Percent Ratio
Opt 1 100 90 30 50.6 8.7 2.56
Opt 2 100 90 30 48.1 8.0 2.35
Opt 3 100 90 30 43.9 6.8 2.00
1 1 100 90 30 36.8 8.5 2.50
2 8 250 90 30 38.8 7.4 2.18
4 3 250 90 60 52.3 6.3 1.85
8 6 250 30 60 40.7 6.3 1.85
5 9 100 30 30 44.4 6.2 1.82
9 5 175 60 45 36.5 6.1 1.79
7 10 100 30 60 37.6 6.1 1.79
3 4 100 90 60 53.6 5.9 1.74
11 7 175 60 45 59.2 5.9 1.74
6 2 250 30 30 40.9 5.5 1.62
10 11 175 60 45 30.9 5.2 1.53
1.000
0.800
0.600
0.400
0.200
0.000
30
37.5
45 100
B: Feed Rate (kg/hr) 137.5
52.5 175
212.5
60 250 A: G-Force (G's)
Figure 4.20. Desirability Surface for Falcon tests with 50-micron feed size.
All three tests produced very high grades, affirming the optimum conditions,
predicted by the desirability surface in Figure 4.20. The rare earth and barium
recoveries double that of the mass recovery, pointing toward the same preferential
separation as the heavy liquid work. The combined grade of these products is also
approximately double that of the feed. Concentration ratios for all tests are given in
Table 4.12.
69
ytilibarise
D
|
Colorado School of Mines
|
4.5 Bench Flotation
As shown in Table 4.13, rare earth grade remained the same, but recovery
improved for the gravity preconcentrate flotation test as compared to the ground ore
test. Calcium recovery was similar, but the grade was reduced.
Table 4.13. Bench Flotation Results: TREE and Calcium Grade and Recovery
TREE Ca
Grade TREE Recovery Ca Grade Recovery
Test (Percent) (Percent) (Percent) (Percent)
P = 45 μm 30 77 9 10
80
Gravity Con 30 82 5 10
Although the gravity preconcentrate yielded a similar grade and higher recovery
for the flotation unit operation, it must be noted that the current gravity recovery is only
50%, meaning overall recovery is only 41%. To eclipse the reported flotation
performance from Molycorp, the process must generate a combined gravity and
flotation recovery of greater than 65%. At this point, it does not. However, looking at the
stepwise recovery and grade from each pass through the Falcon (given in Figure 4.21),
it can be estimated that a sufficient recovery may be produced at an improved grade
with three more passes through the Falcon. This estimation is a linear forecast of the
average grades and recoveries of the first three steps.
Figure 4.21. TREE Grade vs Recovery of Falcon Concentrates.
70
|
Colorado School of Mines
|
CHAPTER 5: ECONOMIC ANALYSIS
Two separate economic models were developed: one utilizing a froth flotation
circuit for concentration and one using gravity preconcentration followed by a froth
flotation circuit. The features of each are given in Table 5.2. Both conceived plant
models treat 400,000 tonnes per year of ore, 8% of which is rare earth oxides. They
have the same comminution circuit in front of the concentrator containing a jaw crusher,
SAG mill, ball mill, and regrind mill. The flotation circuits use 18 rougher, 18 cleaner,
and 36 scavenger cells. Both models’ concentrate products are assumed to be
upgraded by cleaner flotation to 60% REO while keeping the same calcium contents
from the rougher flotation. A hydrochloric acid leach is the next step, where it is
assumed that the acid completely reacts with the calcium in the ore (from calcite),
according to the reaction
.
Equipment sizes and costs were based off those given in the 1,000 tonne/day
Flotation Mill model in CostMine 2014 – the closest model in magnitude to a 400,000
tonne/year operation. Reagent, labor, and energy costs were also sourced from
CostMine or estimated from vendor information. Capital expenses were derived from
the formulas given by Mular. [51] Recent rare earth oxide prices were obtained from the
Metal Pages website. The revenues, profits, and expenses in the section are pre-tax
values.
The first model uses a circuit of flotation cells to recover 65% of the REO content
of the ore (which made up 8% of the total ore body) using a fatty acid collector. The
mine produces 400,000 tons/year of ore. The recovered 20,800 tonnes of rare earth
oxides generate annual profit of $69,390,000 using a $4,000/tonne REO price. The
second model places a series of shaking tables and centrifugal concentrators in front of
the froth flotation circuit, this time utilizing a hydroxamic acid collector. The same
amount and grade of ore is mined but, reflecting the improvement shown by the bench
flotation tests, this design simulates 82% REO recovery. The increased production
revenue and lowered reagent cost (mainly from the acid) yields an annual profit of
71
|
Colorado School of Mines
|
CHAPTER 6: CONCLUSIONS
Cursory microflotation studies showed that although preferential selectivity using
hydroxamate as compared to oleate collectors was suggested by single-mineral
microflotation studies, the same result was not achieved when using an ore mainly
comprised of those minerals. Typical flotation depressants were unable to improve the
selectivity in the system under the conditions explored. While ammonium lignin
sulfonate depressed calcite, it also depressed bastnaesite, leading to no appreciable
change in grade. Sodium silicate raised the pH too high for hydroxamate adsorption.
Copper (II) nitrate addition to the system had no effect on the hydroxamate flotation.
Heavy liquid analysis and gravity concentration trials have shown that the
difference in specific gravities of the minerals in this ore can be exploited to effect a
separation. The calcium-containing minerals (predominately calcite) floated at higher
fluid densities, while the rare earth minerals always sank. This effect was also seen in
the Falcon Concentrator, with the best results (51% TREE recovery) showing that the
longest-ground, most-liberated samples had the highest concentration of rare earths
(8.7% TREE). Qualitative demonstrations on the shaking table suggest it could be used
once the proper conditions have been determined.
Magnetic separation was briefly explored. Barite rejection from the concentrate
was seen in the few tests performed, with an average barium recovery to the magnetic
fraction of 3.1%. However, the mass recoveries of the magnetic fraction averaged
around 7% and showed little change in TREE grade, so no further testing was
performed.
Comparative flotation showed that a 30% TREE, 9% calcium concentrate could
be produced at 77% TREE recovery using hydroxamic acid. Addition of a gravity
preconcentration step reduced the calcium content to 5% while obtaining TREE grades
and recoveries of 30% and 82%, respectively. While flotation of the gravity feed
outperformed flotation of the ground ore feed, the combined process underperformed.
Increasing the number of passes through the Falcon from three to six was estimated to
generate a sufficient concentrate.
76
|
Colorado School of Mines
|
CHAPTER 7: SUGGESTIONS FOR FUTURE WORK
Future work should focus on size-classified centrifugal concentration, to see if
splitting the feed into a coarse and fine fraction would increase overall recovery. The
optimum slurry density was not determined in this work. Plants especially conscious of
their water balance may wish to investigate that factor with a more robust experimental
setup that is resistant to sanding.
Magnetic separation showed evidence of barite rejection, but at uninspiring mass
recoveries. If the mass recovery was improved, perhaps with more passes through the
WHIMS, and it was shown that barite is truly rejected in substantial quantities, flotation
depressant requirements could be reduced.
A combination of gravity, magnetic, and flotation unit operations should also be
investigated to improve plant throughput and reduce flotation reagent consumption.
Gravitational preconcentration operations have shown the ability to upgrade the rare
earth content of the flotation feed, but must be optimized to do so without sacrificing
overall recovery. Together these operations would reduce throughput in the flotation
circuit, lower reagent costs, and improve overall plant grade and recovery.
Once the concentration of the ore by the above methods has been completed,
more surface chemistry analysis should be performed. The zeta potential and
adsorption density of the ore should be investigated before and after concentration, to
determine the effect of reducing the size and removing the calcite. The interaction of the
collectors and depressants on the fine-sized, calcite-free sample could inform on future
reagent selections.
78
|
Colorado School of Mines
|
APPENDIX C: EXPERIMENTAL DATA
Table C.1. Experimental data for microflotation tests
Feed (Calculated) Concentrate Tailings
Depressant Collector Grade (%) Mass Grade (%) Recovery (%) Mass Grade (%) Recovery (%) Balance (%)
Test Dose (M) Type Dose (M) Type Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE Mass Ca Ba TREE
9 1.00E-05 Sodium Oleate 10 33 3 0.48 9 45 3 48 40 66 51 0.44 12 20 3 44 52 26 41 92 92 92 92
1 5.00E-05 Sodium Oleate 13 22 3 0.76 13 22 3 76 77 79 79 0.11 12 16 2 11 10 8 8 87 87 87 87
11 1.00E-04 Sodium Oleate 11 30 3 0.69 11 34 3 69 68 78 73 0.21 12 17 2 21 22 12 17 90 90 90 90
16 1.00E-04 Hydroxamic Acid 14 19 3 0.68 13 23 3 68 64 86 67 0.19 16 1 4 19 23 1 20 87 87 87 87
12 3.00E-04 Hydroxamic Acid 13 24 3 0.84 13 24 4 84 87 81 87 0.07 8 34 2 7 4 10 4 91 91 91 91
10 5.00E-06 5.00E-05 Sodium Oleate 12 26 3 0.69 12 29 4 69 69 77 77 0.22 12 16 2 22 22 14 14 91 91 91 91
2 1.00E-05 5.00E-05 Sodium Oleate 11 33 3 0.63 10 39 3 63 59 74 69 0.30 12 21 2 30 34 19 24 93 93 93 93
Ammonium
5 2.00E-05 5.00E-05 Sodium Oleate 13 24 3 0.50 12 26 4 50 50 54 56 0.41 13 22 3 41 41 37 35 91 91 91 91
Lignin
17 5.00E-06 1.00E-04 Hydroxamic Acid 12 26 3 0.83 12 26 3 83 85 83 88 0.12 10 26 2 12 10 12 7 95 95 95 95
Sulfonate
19 1.00E-05 1.00E-04 Hydroxamic Acid 12 26 3 0.83 12 26 3 83 84 84 87 0.13 11 23 2 13 12 12 9 96 96 96 96
13 2.00E-05 1.00E-04 Hydroxamic Acid 11 31 3 0.74 11 32 3 74 75 77 78 0.12 10 24 2 12 11 9 8 86 86 86 86
6 100 5.00E-05 Sodium Oleate 12 29 3 0.70 12 34 3 70 71 81 69 0.22 11 15 3 22 21 11 23 92 92 92 92
3 300 5.00E-05 Sodium Oleate 12 29 3 0.71 12 31 3 71 70 76 68 0.13 12 17 4 13 14 8 16 84 84 84 84
Sodium
8 500 5.00E-05 Sodium Oleate 11 28 3 0.74 11 30 3 74 74 80 71 0.17 12 19 3 17 17 11 20 91 91 91 91
Silicate
18 100 1.00E-04 Hydroxamic Acid 11 33 3 0.70 11 35 3 70 70 74 71 0.25 11 27 3 25 25 21 24 95 95 95 95
(mg/L)
20 300 1.00E-04 Hydroxamic Acid 13 25 3 0.58 13 25 3 58 60 58 54 0.38 12 25 3 38 36 38 42 96 96 96 96
14 500 1.00E-04 Hydroxamic Acid 12 26 3 0.33 12 30 3 33 31 39 30 0.54 13 23 3 54 56 48 57 87 87 87 87
4 5.00E-07 5.00E-05 Sodium Oleate 12 26 3 0.65 13 28 4 65 66 70 71 0.16 11 18 2 16 15 11 10 81 81 81 81
7 1.00E-06 Copper 5.00E-05 Sodium Oleate 11 29 3 0.53 11 35 3 53 49 65 57 0.41 12 20 3 41 45 29 37 94 94 94 94
21 5.00E-07 Nitrate 1.00E-04 Hydroxamic Acid 11 31 3 0.75 11 30 3 75 78 73 81 0.18 9 35 2 18 15 20 12 93 93 93 93
15 1.00E-06 1.00E-04 Hydroxamic Acid 11 32 3 0.85 11 32 3 85 85 86 85 0.03 12 21 3 3 3 2 3 88 88 88 88
Table C.2. Experimental data for WHIMS tests
Feed (Calculated) Magnetic Nonmagnetic
P80 Field Strength Mass Grade (%) Mass Grade (%) Recovery (%) Mass Grade (%) Recovery (%) Balance (%)
Standard Run (µm) (Gauss) (g) Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE Mass Ca Ba TREE
5 1 762 5000 50 12.7 6.7 2.1 4.4 10 4 2.4 8.8 6.9 5.2 9.9 44 13 7 2.1 88 90 92 87 97 97 97 97
4 2 50 5000 50 15.8 9.8 4.8 1.4 8 2 4.9 2.8 1.4 0.6 2.9 47.2 16 10 4.8 94 96 97 94 97 97 97 97
6 3 762 10000 50 11.0 5.5 2.2 8.4 11 3 3.7 16.8 16.8 9.2 28.2 41 11 6 1.9 82 82 90 71 99 99 99 99
1 4 144 7500 50 16.8 9.7 3.7 2 11 2 3.8 4.0 2.6 0.8 4.1 48.3 17 10 3.7 97 98 100 96 101 101 101 101
3 5 50 10000 50 15.9 8.5 4.9 3.3 14 2 8.4 6.6 5.8 1.5 11.4 46.2 16 9 4.6 92 93 97 88 99 99 99 99
2 6 144 7500 50 16.8 10.6 6.1 2.3 12 3 9.6 4.6 3.3 1.3 7.3 47 17 11 5.9 94 95 97 91 99 99 99 99
Table C.3. Experimental data for bench flotation tests
Feed (Calculated) Concentrate Tailings
Depressant Collector Mass Grade (%) Mass Grade (%) Recovery (%) Mass Grade (%) Recovery (%) Balance (%)
Test Dose (M) Type Dose (M) Type (g) Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE Mass Ca Ba TREE
Ore Ammonium 500 14 11 6.1 78 9 10 30 16 10 14 77 405 15 11 1.5 81 87 82 9955 97 97 97 97
1.50E-03 1.00E-02 Hydroxamic Acid
Gravity Con Lignin Sulfonate 500 12 15 8.9 127 5 12 29 25 10 20 82 347 15 16 1.6 69 84 74 6209 95 95 95 95
98
|
Colorado School of Mines
|
Table C.4. Experimental data for Heavy Liquid tests
Feed (Calculated) Floats Sinks
Mass Grade (%) Mass Grade (%) Recovery (%) Mass Grade (%) Recovery (%)
Test P80 (µm) Specific Gravity (g) Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE
1 50 2.95 5.0 9 8 4.8 1.13 11 0.4 0.14 23 34 1.3 0.8 1.83 5 18 10.5 37 25 96 98
2 50 2.90 5.0 13 7 4.2 1.73 24 2 0.77 35 79 11 7.6 1.72 3 16 9.4 34 10 88 92
3 50 2.80 5.0 11 6 3.3 1.48 15 0.5 0.10 30 45 2.6 1.0 2.44 7 11 5.9 49 35 94 98
4 50 2.70 5.0 10 10 6.0 1.24 13 0.2 0.02 25 36 0.6 0.1 2.76 10 16 9.4 55 62 99 100
5 144 2.95 5.0 11 9 4.4 1.47 15 1.0 0.56 29 47 3.9 4.3 2.01 9 18 9.2 40 38 96 96
6 144 2.90 5.0 9 7 3.5 1.54 13 0.5 0.21 31 50 2.4 2.2 2.05 7 15 7.1 41 36 97 98
7 144 2.80 5.0 10 11 5.3 1.65 12 0.2 0.114 33 46 0.7 0.8 2.21 9 21 10.5 44 46 99 99
8 144 2.70 5.0 14 11 4.9 1.07 9 0 0.00 21 16 0.0 0.0 3.10 14 16 7.2 62 70 99 100
9 762 2.95 5.0 12 6 3.2 1.92 14 1.0 0.56 38 51 7.0 7.5 2.13 7 11 5.8 43 28 86 86
10 762 2.90 5.0 14 10 5.0 2.21 17 1.0 0.77 44 56 4.5 7.3 2.31 12 20 9.4 46 41 95 93
11 762 2.80 5.0 13 9 4.1 1.67 12 0.4 0.17 33 34 1.6 1.5 2.59 12 16 7.1 52 53 97 97
12 762 2.70 5.0 14 10 5.2 1.19 9 0.3 0.06 24 16 0.7 0.3 3.47 16 14 7 69 83 99 100
Middlings
Mass Grade (%) Recovery (%) Balance (%)
Test P80 (µm) Specific Gravity (g) Ca Ba TREE Mass Ca Ba TREE Mass Ca Ba TREE
1 50 2.95 1.11 13 0.7 0.23 22 40 2.3 1.3 81 81 81 81
2 50 2.90 0.72 8 0.3 0.07 14 11 0.7 0.3 83 83 83 83
3 50 2.80 0.51 19 2.0 0.39 10 20 3.6 1.4 89 89 89 89
4 50 2.70 0.35 3 0.1 0.00 7 2 0.0 0.0 87 87 87 87
5 144 2.95 0.87 8 0.2 0.02 17 15 0.5 0.1 87 87 87 87
6 144 2.90 0.71 8 0.1 0.01 14 14 0.2 0.0 86 86 86 86
7 144 2.80 0.52 6 0.0 0.00 10 7 0.0 0.0 88 88 88 88
8 144 2.70 0.36 25 2.0 0.20 7 15 1.4 0.3 91 91 91 91
9 762 2.95 0.48 22 4.0 1.80 10 20 7.1 6.1 91 91 91 91
10 762 2.90 0.16 10 1.0 0.26 3 2 0.3 0.2 93 93 93 93
11 762 2.80 0.33 24 2.0 0.90 7 13 1.5 1.6 92 92 92 92
12 762 2.70 0.03 8 1.0 0.30 1 0.4 0.1 0.0 94 94 94 94
Table C.5. Experimental data for Deister table tests
Feed (Calculated) Concentrate Tailings
Mass Grade (%) Mass Grade (%) Recovery (%) Mass Grade (%) Recovery (%)
Test (g) Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE
1 Recoveries correspond to 500 g feed 500 15 11 3.8 26 8 21 14 5 3 10 19 127 14 14 3.6 25 24 32 24
2 Recoveries correspond to 500 g feed 500 15 11 3.5 53 12 16 7.8 11 8 16 23 180 14 14 3.4 36 33 47 35
3 Recoveries correspond to 500 g feed 500 15 11 3.8 63 9.3 20 9.2 13 8 24 31 160 14 14 3.4 32 30 43 29
4 Feed was Tailings from Test 3 (266 g) 266 16 8 2.4 25 13 14 5.4 9 7 17 21 32 15 10 3.2 12 11 17 16
5 Feed was Middlings from Test 3 (160 g) 160 14 12 3.5 6 17 7.3 2.5 4 4 2 3 120 14 13 3.5 75 74 78 76
6 Feed was Concentrate from Test 3 (63 g) 63 8 23 12.4 18 4.7 26 20 29 18 32 46 0 0 0 0.0 0 0 0 0
4* Recoveries correspond to initial 500g feed of Test 3 500 15 11 4.0 25 13 14 5.4 5 4 6 7 32 15 10 3.2 6 7 6 5
5* Recoveries correspond to initial 500g feed of Test 3 500 15 11 4.0 6 17 7 2.5 1 6 3 3 120 14 13 3.5 24 22 28 21
6* Recoveries correspond to initial 500g feed of Test 3 500 15 11 4.0 18 5 26 20 4 2 12 25 0 0 0 0.0 0 0 0 0
Middlings
Mass Grade (%) Recovery (%) Balance (%)
Test (g) Ca Ba TREE Mass Ca Ba TREE Mass Ca Ba TREE
1 Recoveries correspond to 500 g feed 314 16 9 3.0 63 67 51 50 93 93 93 93
2 Recoveries correspond to 500 g feed 265 17 7.4 2.8 53 58 37 41 100 100 100 100
3 Recoveries correspond to 500 g feed 266 17 6 2.7 53 60 31 38 98 98 98 98
4 Feed was Tailings from Test 3 (266 g) 201 17 6 1.9 76 78 63 60 97 97 97 97
5 Feed was Middlings from Test 3 (160 g) 24 15 11 3.4 15 16 13 15 94 94 94 94
6 Feed was Concentrate from Test 3 (63 g) 39 8.9 22 9.0 62 73 59 45 90 90 90 90
4* Recoveries correspond to initial 500g feed of Test 3 201 17 6 1.9 40 47 22 19 93 93 93 93
5* Recoveries correspond to initial 500g feed of Test 3 24 15 11 3.4 5 5 5 4 93 93 93 93
6* Recoveries correspond to initial 500g feed of Test 3 39 9 22 9.0 8 5 16 18 93 93 93 93
99
|
Colorado School of Mines
|
Table C.6. Experimental data for Falcon tests
Feed (Calculated) Concentrate 1 Concentrate 2
Feed Rate Mass Grade (%) Mass Grade (%) Recovery (%) Mass Grade (%) Recovery (%)
Run G-Force (G's) P80 (µm) (kg/hr) (g) Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE
Opt 1 100 50 30 1278 14 8 5 87 16 16 9.5 7 8 13 14 86 15 14 8.3 7 7 12 12
Opt 2 100 50 30 1199 14 8 4 93 14 14 9.3 8 8 14 19 85 16 14 8.2 7 8 13 15
Opt 3 100 50 30 1214 12 7 3 101 11 11 7.2 8 7 13 18 94 14 13 7.1 8 9 14 16
1 100 50 30 1213 14 9 5 89 12 16 10.6 7 6 13 17 80.9 11 13 7.3 7 5 10 10
2 250 144 30 1194 17 10 4 130 15 11 6 11 10 12 15 68.1 16 11 5.9 6 5 7 8
3 250 50 60 1097 13 7 4 128 13 12 7.1 12 12 19 23 89.1 16 14 7.2 8 10 15 16
4 100 50 60 1150 16 8 4 121 14 11 6.1 11 9 14 18 134.6 17 12 5.9 12 13 18 19
5 175 77 45 1160 15 9 4 119 15 13 7.2 10 10 15 17 91.5 17 14 7.2 8 9 13 13
6 250 144 60 1272 15 8 4 103 13 11 6 8 7 11 11 109.8 16 13 7.1 9 9 14 14
7 175 77 45 1179 9 5 2 102 10 8 5.8 9 9 15 21 88.4 13 10 6 7 10 16 19
8 250 50 30 1142 17 9 5 92 17 14 9.2 8 8 12 16 89.5 18 13 7.2 8 8 11 12
9 100 144 30 1270 17 9 4 192 17 13 7.1 15 15 21 25 111.6 13 9 4.8 9 7 9 10
10 100 144 60 1177 16 9 4 110 15 14 7.3 9 9 14 15 104.6 16 13 7.1 9 9 12 14
11 175 77 45 1230 14 7 4 100 13 10 6 8 8 11 12 94.2 10 7 3.7 8 6 7 7
Concentrate 3 Tailings
Feed Rate Mass Grade (%) Recovery (%) Mass Grade (%) Recovery (%) Balance (%)
Run G-Force (G's) P80 (µm) (kg/hr) (g) Ca Ba TREE Mass Ca Ba TREE (g) Ca Ba TREE Mass Ca Ba TREE Mass Ca Ba TREE
Opt 1 81 16 14 8.2 6 7 11 11 806 13 6 3.3 63 60 47 45 83 83 83 83
Opt 2 67 15 11 5.9 6 6 8 9 715 14 6 2.4 60 59 46 37 80 80 80 80
Opt 3 71 16 12 5.9 6 8 10 10 605 12 5 1.9 50 48 35 28 72 72 72 72
1 100 90 30 71.2 15 14 7.2 6 6 9 9 794.1 15 7 3.5 65 68 53 49 85 85 85 85
2 250 30 30 44.5 19 9 3.6 4 4 3 3 389.7 17 9 3.6 33 33 31 27 53 53 53 53
3 250 90 60 75.8 15 9 3.8 7 8 8 7 556.4 12 5 2.2 51 47 34 31 77 77 77 77
4 100 90 60 123.5 16 11 5.8 11 11 15 18 624.6 16 6 2.1 54 55 41 32 87 87 87 87
5 175 60 45 96.7 14 9 3.7 8 8 9 7 605.6 15 7 3.5 52 52 42 42 79 79 79 79
6 250 30 60 105.3 16 10 5.8 8 9 11 11 633.3 15 6 3.4 50 50 38 39 75 75 75 75
7 175 60 45 85.4 16 11 5.9 7 12 17 18 746.3 8 3 1.08 63 54 40 29 87 87 87 87
8 250 90 30 86.7 14 10 5.8 8 6 8 10 698.9 17 8 3.4 61 62 53 46 85 85 85 85
9 100 30 30 86 17 10 5.8 7 7 7 9 651.1 18 8 3.3 51 54 45 39 82 82 82 82
10 100 30 60 94.5 15 9 3.7 8 8 8 7 603.8 16 8 3.6 51 52 44 41 78 78 78 78
11 175 60 45 88.1 15 9 5.8 7 8 9 11 789.1 14 7 3.5 64 66 60 57 87 87 87 87
100
|
Colorado School of Mines
|
rule, the backfill mortar is not taken into consideration as part of the static calculation
of the tunnel, so there are no special requirements as regards its final compressive
strength.
To achieve a good pumping of the backfill material, the mortar should show a
sufficient flowability. When backfilled, a portion of the mixing water in the mortar
will be released to the surrounding rock which activates the grain structure of the
mortar as supporting medium. The amount of released water depends on the rock
structure. But the release of filtration water can also result in a loss of volume of the
backfill material in the annular gap. The extent of the loss of volume should be
limited by a possible low water / high solids content in the backfill mixture.
A high compactness of the ground is reached by the addition of fine-grained
components such as fly ash. For a quick stabilization of the granular structure, cement
is usually used as the bonding agent. Retarding and stiffening behavior of the mortar
must be adjusted in such a way that the mortar in the injection lines can still be easily
injected even after longer TBM downtimes in order to minimize necessary flushing
and cleaning measures of the injection lines.
The grouting of the annular gap can be realized through openings incorporated
in the segments or simultaneously with the TBM advance by injection lines in the
tailskin (see Figure 3.4).
|
Colorado School of Mines
|
manently displayed. By adjusting the weight at the start and end of TBM advance, the
quantity of mortar injected can be calculated and displayed.
Minimum / maximum injection pressure can be controlled by pressure level
monitoring. The injection points in the tailskin are equipped with pressure transducers
to measure the pressures in the individual grout lines. The disadvantage of this
method is that a small drifts in mix consistency, for example caused by a stiffening of
the grouting material, a higher pressure will be present in the annular gap than will be
effectively measured.
Since the pressure level monitoring takes precedence over the volume control,
it is possible that the injection process at the nozzle is cut off even though the annular
space is not yet completely filled and the maximum pressure not achieved in the entire
annular gap. Therefore, the volume of grouting mortar supplied should not adversely
affect the flow / pressure characteristics substantially during the grouting process.
Data relating to the injected volume and pressure of the individual injection points can
be taken fiom a display screen in the TBM control cabin.
The pressure in the annular gap can permanently be controlled within the
adjustable limit values for the switch-on / off pressure of the grout pumps. If required
the operating data of the grout injection can be stored by means of a so-called process
data acquisition and visualisation and is thus available at any time.
Figure 6.2 shows a schematic diagram of grouting including all necessary
components for the backfilling process up to the injection points at the end of the
tailskin.
|
Colorado School of Mines
|
The two tanks for components A & B are located on the TBM backup.
From the grout injection unit located on the backup component A is pumped via a 30
mm diameter injection line to the tailskin. With an intended minimal delay of 60
seconds the accelerator is pumped via a 6 mm diameter pipe to a valve unit. From
there it flows through a separate hose inside the 30 mm diameter injection line
towards the injector where the two components are mixed.
Both components are transported by means of electrically driven eccentric
screw pumps. In each injection line a flow-meter and a pressure gauge is installed.
The pressure gauge in the component A line is used for processes control while the
pressure of component B is only displayed on the control panel. The mixed backfill
material penetrates into the annular gap after travelling a short distance inside the
tailskin. To avoid dripping of component B upon shutdown of the accelerator pump a
directional control valve is installed in the B-component line just before entering into
the tailskin.
The two component grouting system can be operated in either automatic,
semiautomatic or manual mode. For two component grouting generally the automatic
mode is used whereby the component ratio is preset and the flow rate of the grout is
controlled by the advance rate of the shield.
The manual mode is a service mode for cleaning. In manual mode each line
can be chosen separately and each pump can be adjusted individually.
In semiautomatic mode the percentage of component B is programmed indi-
vidually for each grout line.
After the grouting mode is selected, the sequence of operation for two compo-
nent grout is as follows:
1. Start of excavation and advance cycle
2. Injection of backfill component A starts with 1
3. After approximately 60 seconds (or as set on the grout injection unit)
injection of component B commences
4. Continuous backfill with the mix of components A and B
|
Colorado School of Mines
|
The comparison of the grouting efficiency with ETAC and traditional mortar
showed that the ETAC system would have had to be modified to allow grouting
through injection lines in the tailskin.
The conventional backfill system as installed in the project Botlekspoortunnel
was a fully integrated system and the contractors were more experienced in using this
system. A comparison between the two grouting systems was difficult and was later
regarded as not suitable.
On the tunnel projects West Side CSO Portland and Metro Line 1 Naples the
customers requested for a two component grouting system as an alternative to
standard solutions to achieve the following goals:
a) Quicker bedding of the segment ring
b) No cleaning of the grout lines in case of downtimes and thus enabling a
quicker start up of the TBM advance
c) Exact control of grouting quantities
d) Cost savings for grout pumps
Three competitive dynamic two component grouting systems were developed
as alternatives to conventional grouting systems using single component grouts
(hydraulically hardened h e m ortars). These systems are discussed in the thesis. One
of these systems, the low pressure dynamic grouting system will be highlighted as a
preferred solution fiom the technical and handling point of view.
The low pressure two component backfill system developed by Herrenknecht
AG differs in two major points fiom the ETAC system:
1. The design and arrangement of the grout lines for backfilling of the
annular gap.
2. The grouting process itself.
1. Design and arrangement of grout lines
The design of the ETAC grouting system is characterized by injection lines
welded on the outside of the shield as shown in Figure 7.5.
|
Colorado School of Mines
|
frequently found in real TBM applications. Different types of mixing conditions were
tested, i.e. different injector types and different mixing locations.
In real applications supply of the two components has to be cut off in stages to
avoid any hardening of the grout in the mortar mixing line when interrupting the
grouting process.
Start-stop-tests were carried out to identify the necessary delay between cut-
off of the base mix supply and supply of the accelerator. Based on the observations in
the mixing tests, the delays were selected.
For emergency stops (i.e. supply of both components cut off at the same time)
re-start tests were undertaken.
Tests for breaking up blockages in the grouting line have been performed. For
both tests a blockage of mixed mortar, approximately 15 cm long, was plugged into
the end of the mixing line. In a first test a shutdown period of 14 h and in a second
test a shutdown period of more than 60 h were simulated.
Results of test promam
The calibration tests showed that the piston started moving if the
chamber pressure exceeded 1.9 bar. See Figure 7.9. This value may be
exceeded in the beginning because of the difference between adhesion
and sliding friction. Also the value is slightly dependent on =locity and
to a lesser extent on time.
|
Colorado School of Mines
|
simulated. It was found that there is no interdependence of flow rate and homo-
geneity.
The suitability of the two component system referring to clogging and
cleaning behavior was tested under realistic conditions with the result that staged
start-up and shut-down provide good means to minimize the risk of cloggirig.
7.2.3.3 Active Mortar Tests with products from Mapei
The same tests and test set-up for the products from supplier Mapei SpA of
Italy were conducted in order to specie the requirements for the grouting components
to be used in Portland.
Mapei two component mortar
The mix with products from Mapai is composed as follows:
Formula # 1 Formula #2
Water 870.0 kg 842.0 kg
Bentonite 34.7 kg 35.8 kg
Cement 23 1 O. kg 238.0 kg
Fly ash 46.7 kg 101.0 kg
Mapeistab 2.3 kg 2.4 kg
Mapeiquick 51.0 kg 83.0 kg
Mapeistab is a stabilizer that is added to component A (base mortar) to avoid
hydration of cement. Mapeiquick is the accelerator mixed with component A in the
mortar line.
Herrenknecht AG received two different mixes from Mapei SpA using
different cements. The main difference between the two mortar formulae was the
content of fly ash and water. The other ingredients needed to be adjusted in order to
maintain certain properties of the fresh and hardening mortar.
To gain a first impression of the properties of the grouting material of the two
different formulas some hand-mixing tests were performed. These quantities were
rather small and a laboratory style of mixer was used. Ten to fifteen minutes after the
|
Colorado School of Mines
|
described in chapter 7.2.3. The active mortar tests showed that a homogenous two
component mix is achieved.
On site experience gained with this injector type during the EPB excavation on
the project Metro Line 1 Naples showed that this design is disadvantageous because
material is accumulated below the silicone rings resulting in improper accelerator
distribution in the mix. Material accumulations at the comers of the sealing surface
caused the silicone rings not to seal properly. Blockages in the injection lines occurred
because base mix could enter the injector.
7.2.4.2 Step No 2: Iniector with one bore cross section and one packer
The injector design as described in step 1 was improved. The objective was to
avoid injection material accumulating below the packers and that maintenance of the
injector via a lock is possible.
The experience gained with the initial injector design showed that after a short
application time (over about 6 meters of TBM advance) the rubber lip at the injector
was damaged with the result that the packer was clogged with material. This led to
blockage of the injection line and complete destruction of the silicone seals resulting
in blocked openings. Design changes were made: a protecting cage consisting of three
steel rods was welded to the injector head to avoid loss of silicone rings. This
modification led to an increased life span of the nozzle; 37 rings were erected with the
same nozzle. During the visual inspection of the nozzle the silicone rings were still
present as shown in Figure 7.17.
|
Colorado School of Mines
|
Behavior during restart of pumping after long standstill (clogging of the
injector?)
Visual examination of the mixing ratio by using Ultra White Cement for
component A and graphite powder (4 pm) for component B.
7.2.6.1.1 Test set-ur, and test material
For technical reasons the annulus was simulated by means of a steel vessel, i.e.
it did not have the same curvature as the segment ring.
The setup consisted of the following components (see Figure 7.3 1):
Pressure vessel for washout tests with running water:
Length of 600 mm
300 mm diameter.
Port for water, pressure gauge and pressure regulating valve
Port for 2-component backfill mix
Pressure vessel 300 mm diameter with lateral openings for simulation
of water flow.
Mixing section (ID= 30 mm)
Suspension line (ID= 30 mm) for component A
Injection hose (ID==6 mrn) for component B
Two pumps for components A and B
A mixer for blending of component A.
The material for the test comprised a test soil with a grain size distribution
curve similar to the soil found on the project Westside CSO Portland characterized by
a permeable gravel alluvium as shown in Figure 7.32.
|
Colorado School of Mines
|
structure. Whether the mortar has a yield point and, should this be the case, which
yield stress is necessary to overcome the yield point as well as the correspoding
viscosity is important for the evaluation.
The measurement of the density is used to control the quality of component A.
With an adjustment of the density, buoyancy effects can be balanced.
The viscosity is a useful value to characterize the consistency of the grout and
to control the grout penetration. The viscosity generally gives information about the
flow properties of the backfill component.
With information about setting ratio, the stability under atmospheric con-
ditions can be determined. An objective is to avoid a release of water and the segre-
gation of mixtures during injection. An eventual segregation would have a negative
affect on strength and durability of the backfill.
After each of the 11 test runs to be described in this chapter, the two com-
ponent mix was filled into concrete test cube moulds (150 mm x 150 mm x 150 mm)
to determine the unconfined compressive strength (UCS) of the hardened backfill
material after 3, 7 and 28 days. UCS test were performed in the laboratory of
company "Sch6ck Bauteile GmbH", Baden-Baden.
Course of practical testing and test eauipment
Washout resistance tests of two component backfill material were carried out
at Herrenknecht AG in Schwanau between the fh of March 2005 until the lgth of
March 2005.
A test rig consisting of a steel tube and pipework for material transport and
injection was designed and built. See Figure 7.34.
|
Colorado School of Mines
|
The data in Figure 7.37 illustrate the injection of both components. It can be
seen fiom this diagram that at the end of the injection process, the flow of component
B is switched off a few seconds earlier than component A to avoid a clogging in the
mixing section. With the delayed cut-off of component A the mixing section is
flushed with component A. This avoids component B remaining in the line conse-
quently avoiding blockages as a result of hardening grout.
The functionality of the new injector design was verified. The grout showed a
homogenous mix (see Figure 7.37) and no marble-effect indicating poor mixing.
Test 3 - Depressurized vessel, constant water flow, discharge valve closed
For this third test, the components were mixed in two 42 1 batches as per mix
proposed by Condat. The formula (see Table 7.13) is adjusted to permeable soils with
running water conditions.
First of all the flow properties of component A were tested as described in
chapter 7.2.6.1.2. Laboratory tests. A summary of the properties like density, liquid
flow limit, viscosity and settling behavior is presented in Table 7.14 following the
description of the 11 tests.
Before commencing the test concrete test cube moulds (150mm x 150mm x
150mm) were prepared for UCS tests on samples of the two component backfill mix
after 3, 7 and 28 days.
The test tube was filled half with gravel alluvium. A constant water flow
through the depressurized vessel was generated with the discharge valve closed.
|
Colorado School of Mines
|
Test 6 -Pressurized vessel (4 bar), water filled vessel, constant water flow
For tests 6 and 7 new components A and B were mixed as per the mix
recommended by Condat, adjusted to resist washout in running water conditions. The
properties of the component A mix were determined in accordance with chapter
7.2.6.1.2.
Because the operational data for test 6 were registered faulty, it is not possible
to illustrate the test run in form of a diagram. The test was run with an adjusted
constant water flow. Then the vessel was pressurized to 4 bar maintaining a water
flow of 4 to 5 Vmin. The two components were then injected into the vessel and the
pumps were stopped when mix poured from the outlet.
The material at the surface of the sand was liquid. Probe No 4 was taken from
the vessel to determine the unconfined compressive strength of the two component
mix after 28 days.
Test 7 - Pressurized vessel (4 bar), constant water flow, constriction-hose valve 4 bar
The constriction-hose valve was adjusted to 4 bar. Then a constant water flow
of 4.5 to 5 11s was established through the vessel until the complete tank was filled
with water. This was recognized by water flowing out of the discharge valve. The
vessel was pressurized with 4 bar when the two component injection started. The
water flow stopped shortly after the two components had been injected and the
pressure in the vessel increased to 14.5 bar. The component injection was continued
until the mix poured out of the discharge valve.
The two components formed a thin and relatively liquid, layer as illustrated in
Figure 7.42.
Two cube specimens were taken, No 511 from the discharge valve to test the
UCS after 7 days and No 512 from the vessel to test the strength after 28 days.
|
Colorado School of Mines
|
7.2.6.1.3 General Result of the washout-resistance tests "running water" tests for two
component backfill material
All 11 tests show significant results with regards to the injection of two com-
ponent backfill material under running water conditions. Not all tests runs ended with
a completely filled chamber but in each test, the material intruded into the sand and
formed a filter cake and sealed the soil against water ingress.
Washout resistance and mixing behavior of two component backfill material:
The washout resistance of the two component backfill material could be
verified with each of the accomplished tests (test 2 to test 11) as illustrated in chapter
7.2.6.1.2 experimental runs.
Also the injector types (Figure 7.46 and Figure 7.49) used for the tests showed
good results with regards to homogeneous mixing of the two components. The sealing
ring fitted around the injection openings as shown in Figure 7.46 is pre-stressed to 0.5
bar. The arrangement does however not stop the flow of accelerator instantly once the
pump is switched off. Blockages are the consequence. The design is therefore not
suitable for on site application. The injector illustrated in Figure 7.49 in comparison is
characterized by a robust design as described in the following section.
Durabilitv of used injector type:
Experience was gained on one undocumented test run because an operating
error caused by operating personnel lead to a blockage in the mixing section. During
the course of testing first the pump for the component A was switched off instead of
fmt switching off the flow of component B in order to flush the line with component
A to keep it free from clogging. On TBMs this operating error can be avoided by
presetting of pumps and flow rates. This error was however helpful in order to test the
injector type under extreme conditions. After the blockage in the mixing section
occurred the injector was disassembled. No damage or clogging of the material at the
outlet of the injector was visible. Thus this injector can be regarded as tough and was
|
Colorado School of Mines
|
For the project Brescia with 6 +1 segments altogether 6 inclinometers have
been provided.
The basic idea is the assumption that any deformed ring could be considered
as a chain of rigid bodies linked by joints. Deformations could take place only along
longitudinal joints, to be treated as articulations (courtesy of VMT).
The deformation of the entire ring will be determined by collecting tilt
changes of the inclinometers. The obtained coordinates of the articulations will permit
the calculation of any convergence line. Convergence accrues from the differences
between any two arbitrarily chosen time instances (so-called epochs).
With the help of inclusion and measurements to one or more reference
prism(s), absolute movements can be observed and contribute to the convergence
measurement analysis.
For the project Brescia one tunnel ring (ring number 58) was selected and
equipped with the RCMS. The very first record (one set of inclinometer readings) was
defined as zero measurement. To start the evaluation the 2D co-ordinates of the
articulations are required. With each additional series of inclinometer readings
(epoch) the changes of the inclinations were measured and the same evaluation as
with the zero epoch were repeated.
The comparison with the spatial distances of the zero epoch yielded small
differences, which permitted conclusions about the stability of the ring and also of the
lining in its vicinity. The accuracy of the results increases with the number of
measurements.
The set of displayed dashed lines in Figure 7.51 are the maximum available
convergence lines, with the three diagonals being of prime interest, and among them
the near vertical line S1-S4 being the most important.
|
Colorado School of Mines
|
could be observed by the servo-theodolite of the TBM Guidance System SLS-T. For
the integration into the Guidance System, a software adaptation was necessary.
Measurements and data recording of the segment reference prism were performed
automatically, after a preliminary system calibration. Subsequently, one can split the
vertical convergences into two separate components "crown settlement" and "invert
lift". Should it happen that both components exhibit the same sign, then a tunnel lift-
up is detected.
The software for the RCMS serves for recording inclinometer measurements.
Output of inclinometer readings with a resolution of 1/1000°
Each sensor individually activateddeactivated by a controller
Single and permanent measurements are possible
Recording of measurements in a database
Recording of all data in log-files
Export to e.g. MS Excel.
For the data evaluation and visualization the program requires the geometry of
the ring type (co-ordinates of the articulation points) and the database of the
measurements of the monitored ring, in this case ring 58.
In the following Table 7.1 5 the main components of the RCMS are described.
Figure 7.52 shows a schematic of the RCMS and its main components.
|
Colorado School of Mines
|
The steps for the analysis of the RCMS data are:
1. Selection of the ring to be equipped with the RCMS
2. Installation of the inclinometers, prism and controller unit
3. Activation of the inclinometers by the controller and automatic recording
of data of prism and inclinometers
4. Definition of reference data record (zero measurement)
5. Automatic recording of data from prism and inclinometers
Steps 1 to 3: For the measurement and data recording an inclinometer was
installed on each of the six segments of the selected ring but not the key. One per-
manent prism was installed in the crown to be able to determine the absolute displace-
ments. After the controller unit was fixed, the convergence lines were defined.
Step 4: The reference data record (zero measurement) was defined before the
recorded data from the RCMS database was recorded. The zero measurement was
data set number 1 of 279 dated February 3,2006,1853: 15 pm.
The RCMS recorded data between February 3,2006 and February 13,2006.
Step 5: The analysis of the data from the permanent prism shows no displace-
ments: no settlement of the crown, no lift of the invert or even a lift-up of the entire
tunnel tube. The absolute vertical displacements of the crown prism are shown in
Figure 7.53.
Figure 7.53 illustrates that over a period of 7 days the segment reference prism
moved from +I .6 mm to -2.6 mm.
With the information gathered from the tilt changes of the inclinometers along
the 26 meters of the test section over a time period of about 7 days the deformations
of the entire ring from the zero measurement to the last measurement (data set 279)
could be determined. The obtained coordinates of the articulations permitted the
calculation of the interesting convergence lines as depicted in Figure 7.54.
|
Colorado School of Mines
|
Comparing the observed very small vertical displacements of only 2 mm to 3
mm (see figure 7.53) of the crown prism with the finally detected vertical
convergence of +1.8 mm the following conclusions can be drawn for project Brescia:
that the crown had settled by about -2.6 mm
that the (almost) vertical convergence line S1 - S4 had increased for
about +1.8 mm, leading to an apparent invert settling of 4.4 mm.
The measuring results and the performed studies indicate that convergences in
the annular gap can be kept to a minimum by using two component grouting material.
However, it cannot be necessarily gathered from these results that they are exclusively
based on the effect of the two component grout. In order to prove this effect a test
series with standard grout would have been necessary under the same tunneling
conditions.
The experiment showed by means of the RCMS that the ring shape and the
injection of the annular gap with two component grout could be kept stable during the
measuring period.
7.3 Two component high pressure annular gap backfilling
In parallel to the low pressure system (max. 16 bar) the development of an
high pressure injector was also considered which could be used to inject accelerator at
a maximum of 120 bar to achieve an even more homogenous backfill material. This
requires however modifications to the grout injection pipe cross section and the in-
jection pumps.
7.3.1 Desim and function of the high pressure backfill svstem
This system works with significantly higher pressures (approx. 120 bar) than
the low pressure injection system with maximum pressures of up to 16 bar.
|
Colorado School of Mines
|
The injection pipe conducts the backfill mix towards the annular gap. In both
flushing lines a movable hollow rod is installed with a gate vale mounted to a
hydraulic cylinder allowing axial movement of the assembly.
The gate valve at both ends has the same diameter as the flushing line. The
middle section of the gate valve is tapered. At the front part of the gate valve in
direction of the annular gap, a notch is milled with an opening for water. Above the
notch an inflatable seal is mounted.
When the rod moves towards the segment ring, the valve closes the pipe
shortly behind the inlet of the injection line. In extended position the valve closes the
grout supply line. In retracted position the valve closes the pipe before the injection of
grout and opens a part of the pipe which leads into the annular gap.
When the injection process is started, the valve is hydraulically retracted
opening the tail pipe. Via the injection line the mixed backfill material is pumped into
the annular gap.
Upon completion of the injection, base mix will be pumped for a short time
through the pipes to avoid remaining mixed backfill material in the pipes to prevent
hardening. When the remaining two component mix is completely pressed out of the
line, the valve extends again and closes the supply to the annular gap. To clean the
pipes, water will be flushed through the hollow rod holding the gate valve. The
openings in the valve provide a connection between flushing and injection lines with
water flowing in the grout line. The water cleans both lines fiom grout.
7.4.2 Development steps of the premix variant
The first modification relates to the concept of the cleaning system. With the
variant described in the chapter "design of the premix variant" the two hydraulic
cylinders necessary for the movement of the valve would have to be small and slim.
Design and customer requirements require that hydraulic cylinders should be
installed. In addition the tubular valve rod should be replaced by a more robust and
|
Colorado School of Mines
|
Metro Line 1 Naples
Two EPB-Shields 6.74 m diameter that are in operation for Metro Line 1 in
Naples will each drive a tunnel length of approx. 4,340 m through clay and tuff. Both
machines are identical, equipped with four injection groups each comprising two
lines, one for conventional grouting and one for two component injection. The
arrangement of the groups A1 to A4 and of the injection lines 1 to 8 is shown in
Figure 9.2.
Figure 9.2 Arrangement of the injection groups and lines for annular gap backfilling
(Lines 2,3,6 & 7 for two components with DN30 and lines 1,4, 5 & 8 for conven-
tional grouting with DN50)
For both machines the operational data of the complete tunnel section have
been analyzed to show how the backfilling of the annular gap with a dynamic
grouting system performed. Data relating to the grout and accelerator flow in lines 2,
3,6 and 7 has been considered.
The average mixing ratio of components A and B in the injection lines has
been determined.
|
Colorado School of Mines
|
The interpretation of the diagram shows that the two components have been
permanently injected as planned through all two component lines 2,3,6 and 7 (groups
A 1, A2, A3 and A4).
For EPB-Shield No 2 the same analysis of operational data has been under-
taken for the first section. Table 9.2 shows the average mixing ratio of component A
& B for a section of about 1,O 15 m.
Table 9.2 Average mixing ratio in two component injection lines 2, 3, 6 & 7 for EPB-
Shield No2 .
Ring 0Flow 0Flow @Flow 0Flow 0Flow 0Flow 0Flow 0Flow
No Comp. Comp. Comp. Comp. Comp. Comp. Comp. Comp.
A B A B A B A B
Line 2 Line 2 Line 3 Line 3 Line 6 Line 6 Line 7 Line 7
["/.I [%I ["/.I [%I ["/.I ["/.I [%I
[%]
10-879 94 6 95 5 100 0 95 5
Figure 9.4 illustrates the two component grouting along the first tunnel section
of EPB Shield No2 from ring 10 to 879.
|
Colorado School of Mines
|
Circle Line 852, Singapore
Grouting the tail void with the two component low pressure system with max.
16 bar was also foreseen for the two EPB-Shields 6.60 m diameter that have been in
use for the excavation of two approx 1,870 m long Metro tunnels for the Circle Line
852 in Singapore. The machines faced geological formations such as marine clay, old
alluvium, sand and weathered granite.
The EPB-Shields were similarly equipped to the TBMs for Naples and Port-
land with four DN30 mm lines for injection of the two components and four oval
grout lines DN50 mm for the injection of conventional single component grout.
Figure 9.5 Arrangement of the injection groups and lines for annular gap backfilling
(Lines 1,4,5 & 8 for two components with DN30 and lines 2,3,6 & 7 for conven-
tional grouting with DN50)
The two TBMs started excavation by the end of December 2004 and end of
March 2005 ~spectively.T he dynamic two component grouting system was active
from the beginning of the tunnel drives. Component A consisting of Cement,
bentonite, plasticizer and water was mixed with Component B with a ratio of 92-93%
A to 7-8% B. Component B is an accelerator.
Based on experience made on site, the circular injection lines 1, 4, 5 and 8 that
were specially installed for the backfilling of the two components haven't been used.
These lines ON30 mm) were considered to be too small for incorporating the injector
|
Colorado School of Mines
|
ABSTRACT
Distributed acoustic sensing (DAS) is a relatively new technology used in many geophysical
applications. Its first notable use for geophysical monitoring includes downhole deployments for vertical
seismic profiling and more recently, hydraulic fracture characterization in unconventional wells. Over the
last decade, DAS has extended into broader seismic applications including structural imaging, and
near-surface surveys for seismic site classification. Critical to these different applications is the availability
of cost-effective methods to acquire data in logistically challenging settings. We propose two novel
procedures spanning two distinct industries: (1) using the low-frequency band of DAS to diagnose
multi-stage hydraulic fractures in upstream oil and gas, and (2) surface-deployed DAS optical-fiber for
low-impact seismic hazard geotechnical surveys.
In 2020, 13 horizontal wells were drilled and completed at the DJ-Postle wellsite, including three wells
with various fiber installation methods (permanent, wireline, and disposable) to evaluate completion design
efficiency. We apply a geomechanical inversion algorithm to constrain fracture widths using low-frequency
DAS (LF-DAS) recorded at an offset well to evaluate the degree of stage isolation in an injection well.
LF-DAS indicates incomplete stage isolation in three of the four analyzed intervals, which is validated with
distributed temperature sensing (DTS) measurements recorded in the injection well. We find
high-frequency in-well DAS measurements are affected by proppant induced erosion and near-wellbore
fractures, preventing reliable diagnostics. Implications of our results support LF-DAS for providing critical
information for in-well diagnostic interpretations to optimize completion efficiency.
We then leverage the versatility and sensitivity to surface-waves of DAS to examine the potential for
using untrenched surface deployments. We acquire continuous DAS data for one hour on a rapidly
deployed fiber array composed of six parallel linear subsections laid directly on the surface with different
fiber-ground contact conditions. We apply ambient interferometry and adopt a simplified
spectral-analysis-of-surface waves (SASW) method to determine the average shear-wave velocity of the top
30 m (VS30). Our methodology results in VS30 estimates for each surface subsection that are consistent
with collocated 1 m-depth trenched cables. The implications of these findings support DAS as a viable
method for non-invasive deployment surface surveys for earthquake hazard assessment.
iii
|
Colorado School of Mines
|
CHAPTER 1
INTRODUCTION
1.1 Motivations and Background
Distributed acoustic sensing (DAS) is a rapidly evolving technology belonging to a classification of
techniques known as distributed fiber-optic sensing (DFOS), including distributed temperature sensing
(DTS), and distributed strain sensing (DSS) [2]. Its first notable geophysical use was in the oil and gas
industry for downhole monitoring (i.e., vertical seismic profiling, continuous microseismic, time-lapse
monitoring) [2][3], and more recently hydraulic fracture characterization in unconventional lateral wells
(e.g.,[4][5]). Within the last decade, the use of DAS has extended into broader seismic applications
including but not limited to carbon capture utilization and storage (CCUS) (e.g., [6]), earthquake and
aftershock monitoring (e.g., [7][8]), ocean bottom surveys (e.g., [9]), and near-surface imaging in dense
urban environments (e.g., [10]).
The aforementioned applications vary widely in their deployment methods, acquisition parameters, and
purpose for the end-user. However, they are linked by the versatility of DAS, and its enabling capability to
deploy ultra-dense seismic networks at a relatively low-cost. The rapid success of DAS in both industry and
academia is arguably the continued demonstration of its practical use in the field. To ensure and accelerate
the use of DAS among a diverse range of scientists and technicians, it is critical to innovate multi-domain,
cost-effective methodologies. Accordingly, this thesis leverages the high spatial-temporal resolution and
exceptional broadband capabilities of DAS to develop two novel methodologies spanning two distinct
industries: (1) using the low-frequency band of DAS (LF-DAS) to diagnose multi-stage hydraulic fracture
treatments, and (2) surface-deployed DAS optical-fiber for low-impact seismic site classification surveys.
Over the last decade, the evolution of hydraulic fracturing completion designs has simultaneously
increased fracture density and lateral well length [11]. Rapid expansion of fracturing efficiency has
outpaced the ability to monitor and constrain the complexity during completion phases, limiting available
information, and ultimately impacting upstream production. DTS and DAS are two mature methods used
for monitoring and diagnosing completion activities in upstream oil and gas. The most common practice
includes the permanent installation (i.e., cemented to formation behind wellbore casing) of multi-mode
and/or single-mode optical fiber to acquire continuous data during the injection of hydraulic fracturing
fluid. However, permanent installation is time consuming and logistically challenging, imposing a
significant cost to operators and service companies.
1
|
Colorado School of Mines
|
It has been demonstrated that the low-frequency band of DAS contain valuable information critical for
characterizing hydraulic fractures [4]. Optical fiber installed in an offset well at some distance away from
an injection well, record low-frequency strain perturbations in the surrounding formation caused by
approaching and intersecting fractures. This application has been primarily limited to qualitative analysis.
Recently, a quantitative approach has been developed by Liu et al. (2021a, 2021b), which constrains the
fracture width at an offset well through inversion of the recorded strain measurements. This quantitative
analysis capability sets the stage to use LF-DAS as a complementary, or stand-alone hydraulic fracturing
diagnostic tool for completion efficiency. Furthermore, use of optical fiber in offset wells does not require
permanent installation. Recent field studies demonstrate temporarily deployed cables can produce LF-DAS
signalswithsimlarqualityaspernanentlyinstalledcables[14][15]. Fiberinanoffsetwellisalsonotexposed
to the risk adverse environment of an injection well, reducing the likelihood of damage and associated cost.
In the last half decade, the popularity of DAS has extended beyond the oil and gas industry, and its
applications have grown in near-surface geophysical studies (i.e., engineering, infrastructure, earthquake
monitoring). The rapid deployment of dense seismic networks has long been a challenge within the greater
seismic community. In recent years, leveraged technology has included portable nodes or geophones, snow
streamers (i.e., hydrophones acting as an array on glaciers), land streamers, and the 2D/3D autojuggie
[2][16]. However, these methods are all limited by their capabilities to be deployed in different
environments. Perhaps the most widely used method for seismic acquisitions, geophones, are difficult to
deploy on hard surfaces such as hard-rock layers or cement and asphalt. Additionally, source distribution
places constraints on the array geometries, further restricting the use of some seismic instruments.
The versatility and high spatial-temporal resolution of DAS provides an economic low-impact
alternative to traditional seismic methods. In recent years, the high sensitivity of DAS to surface waves
(i.e., Rayleigh) has been leveraged for near-surface surveys (e.g., [17][1]). Common among these
configurations is the trenching of the optical fiber (< 1 m) beneath the surface. While this enhances
coupling with the formation to increase signal-to-noise ratio, it also increases the cost and logistics of DAS
deployments. It restricts acquisitions to locations with previously installed fiber (so called “dark-fiber”)
and/or where environmental factors are not a concern. The development and application of low-impact
surveys using DAS (i.e., on the ground surface) would greatly contribute to the advancement of its use
beyond oil and gas, and demonstrate its value in the broader seismic community (i.e., environmental
geophysics, geotechnical surveys).
2
|
Colorado School of Mines
|
1.2 How Does DAS Work?
DAS repurposes optical fibers used for telecommunication into multi-channel seismic arrays [17].
Instead of transmitting information, DAS relies on the principles of time-domain reflectometry to
effectively turn a length of fiber-optic cable into a continuous network of seismic sensors out to tens of
kilometers [2]. The DAS instrument is referred to as an interrogator unit that sends laser pulses into a
connected fiber-optic cable. Strain and temperature variations are recorded along the fiber by measuring
the properties of backscattered light caused by refractive index heterogeneities in the glass fiber core.
Specifically, DAS uses Rayleigh scattering to measure dynamic strain every few meters along the axis of the
fiber. Slight perturbations caused by seismic waves or other vibrations change the distance between
adjacent sections of the fiber with the return signal carrying information of the disturbance.
Unlike traditional sensors that rely on discrete point measurements, DAS is a distributed sensor that
measures strain changes across a length of the fiber known as gauge length, typically between 1-40 m. The
gauge length acts as a differential operator across the spatial axis of the data to convert the measurements
to strain rate equivalent. The spatial axis of the DAS data is reported in so-called ”channels”, with much
finer spacing than gauge length. Effectively, the spatial resolution of DAS is fixed by the gauge length at a
sampling frequency of channel spacing.
1.3 Scope and Outline
In Chapter 2 we develop and demonstrate an effective methodology using LF-DAS data recorded at an
offset-well to diagnose multi-stage hydraulic fracture treatments in an unconventional lateral well. This
novel approach adopts the geomechanical fracture width inversion algorithm presented by Liu et al.
(2021a, 2021b). This study was conducted as part of the larger DJ-Postle Integrated Project, which is the
Reservoir Characterization Projects (RCP) primary field project for Phase XIX research. The project
includes a multi-disciplinary team of geologist, geophysicists, and petroleum engineers to integrate a
comprehensive dataset to evaluate completion efficiency and well communication in the Denver Julesburg
(DJ) Basin, Colorado. The well-site, located 20 miles north of Denver, is owned and operated by Great
Western Petroleum and includes eight parent wells completed in 2018, and 13 children wells drilled and
completed in late 2020. The field provides a robust and unique suite of distributed fiber-optic sensing data,
including permanently installed in-well DAS and DTS, and LF-DAS data recorded at wireline retrievable
fiber and pump down dissolvable fiber offset wells. The proposed diagnostic method has the potential to
complement, or even replace, current in-well diagnostic techniques, at a proportion of the cost.
Chapter 3 develops and demonstrates the ability of surface-deployed DAS cable to record high-fidelity
surface waves (i.e., Rayleigh) and produce robust results used for seismic hazard site classification. Data
3
|
Colorado School of Mines
|
CHAPTER 2
QUANTITATIVE STAGE ISOLATION EVALUATION USING CROSS-WELL STRAIN
MEASUREMENTS
2.1 Introduction
One of the primary objectives during multiple-stage hydraulic fracturing treatments is the efficient
allocation of fluid and proppants into perforation clusters to optimize well performance. The
plug-and-perforation (plug-and-perf) completion design in horizontal unconventional wells rely on complete
isolation between treatment stages to effectively stimulate the surrounding formation. Stage isolation is
achieved when fracturing fluid and proppants exit the wellbore only through perforations at the targeted
interval depths. When stages are not well isolated, fluid and proppant leak from the targeted treatment
interval into one or more preceding intervals.
Incomplete stage isolation is commonly attributed to partially set bridge plugs, plugs set in deformed
casing, and incompatible plug specifications (i.e., size, temperature, pressure ratings) [18]. Significant
degradation of both the casing and bridge plug due to proppant-induced erosion can also lead to
confinement issues [19]. The loss of stage isolation has a significant impact on treatment efficiency, leading
to risks associated with uncontrolled fluid and proppant allocation including damage to the casing and
near-wellbore region, and understimulated fractures in the targeted interval [5, 18–20]. Because well
completion is often the most expensive cost to operators, the improvement of efficient completion design
and diagnostic methods is crucial for optimizing delivery and placement of hydraulic fracturing treatments.
Distributed fiber-optic sensing (DFOS) technology has been increasingly used over the last decade as an
advanced monitoring system during completion and production of unconventional wells [4, 21–24]. DFOS
uses the principles of time-domain reflectometry to effectively turn a length of a fiber-optic cable into a
continuous network of seismic sensors [2]. Strain and temperature variations are recorded along the fiber
by measuring the properties of different backscattered lights, including Rayleigh scattering (DAS),
Brillouin scattering (DSS), and Raman scattering (DTS). DFOS cable can be either installed inside a well
or permanently cemented behind casing, exposing the fiber-optic cable to the effects of fluid flow,
temperature changes, and mechanical deformations in the surrounding formation [24].
Current DFOS diagnostics use high frequency DAS and DTS measurements acquired inside the
injection well (so called “in-well”) during hydraulic fracturing treatments to estimate volumetric fluid
distribution at cluster and stage levels [22, 24, 25]. However, in-well measurements are logistically
challenging and costly, requiring permanent (i.e., cemented to formation) installation in the wellbore.
5
|
Colorado School of Mines
|
Additionally, there is still uncertainty surrounding how best to produce quantitative results with in-well
DAS, as it generally relies on empirical models and assumes a correlation between acoustic energy and fluid
flow. Qualitative interpretation using DTS is relatively simple, however solving transient temperature
behavior requires rather complicated thermal forward modeling to produce quantitative results [24, 26].
Alternatively, DAS cable can be deployed in an observation or offset well to acquire low-frequency
strain measurements. Low-frequency DAS (LF-DAS) applications have been largely restricted to
qualitative and semi-quantitative analysis to characterize far-field fracture geometry (i.e., fracture height,
length, and density) [4, 14]. The development of quantitative methods using LF-DAS is crucial to advance
its use for hydraulic fracture diagnostics. Recently, Liu et al. (2021a, 2021b) demonstrate the application
of a geomechanical-based inversion algorithm to estimate hydraulic fracture widths using low-frequency
strain measured at an offset well. Quantitative analysis in the far-field provides the foundation for using
LF-DAS to evaluate treatment efficiency and optimize completion parameters. Additionally, offset well
deployments do not require permanent installation and are less likely to be damaged during treatment,
substantially reducing the economic risk to operators and service companies.
The objective of this work is to build an approach to diagnose hydraulic fracturing treatment stages
using LF-DAS data recorded at an offset well. Our methodology includes qualitative in-well DTS and DAS
analysis during the hydraulic treatment of four adjacent stages. We then apply the inversion algorithm
proposed by Liu et al. (2021a, 2021b) to the LF-DAS data, constraining far-field fracture widths. Strong
agreement between in-well DTS and LF-DAS inversion results confirm incomplete isolation in three of the
four stages, while in-well DAS is inconclusive. Results demonstrate that low-frequency far-field strain can
be used to confidently diagnose hydraulic fracturing treatments.
2.2 Field Description and Instrumentation
The study area is in the highly productive unconventional play located within the Wattenberg field of
the Denver Julesburg (DJ) Basin, Colorado (Figure 2.1). The DJ-Postle well-site is owned and operated by
Great Western Petroleum and targets four hydrocarbon bearing horizons: The Niobrara A, B, C, and
Codell (Figure 2.2). The Niobrara is a late Cretaceous formation consisting of alternating chalk and marl
units, with production generally best in the fractured, high TOC chalk beds. The general thickness of the
Niobrara formation ranges from 300-400 ft across the Wattenberg field, with individual chalk beds ranging
from 20-50 ft [27]. In the Wattenberg, the Codell is characterized as an impermeable, fine-grained marine
shelf sandstone ranging from 5-20 ft in thickness [28].
The DJ-Postle pad includes 13 horizontal wells drilled and completed in 2020 (Figure 2.2). All wells are
two-mile laterals with varying completion designs (i.e., stage spacing, cluster spacing, proppant volumes).
6
|
Colorado School of Mines
|
Figure 2.1 Map view of the Denver Basin including the Greater Wattenberg Field (in red) and the
approximate location of the DJ-Postle well site (blue star).
The wells were completed by so-called zipper fracking (i.e., hydraulic stimulation of multiple wells in
sequence), with two zipper groups operated by two fracking crews, resulting in a complex dataset with up
to four different wells being treated simultaneously. This study will focus on zipper group one (green wells
in Figure 2.2) which includes seven wells located in the four target formations, completed between
November 16th and November 30th, 2020. Each well ranges between 36-51 hydraulic fracture stages
resulting in 322 stages total during the two-week period.
Continuous data were acquired by three optical-fiber cables deployed along the length of the wellbores
during the two-week operation: (1) a permanently installed (cemented behind the wellbore casing) fiber
located in the Niobrara B well N1B highlighted by the magenta pentagon, (2) a retrievable wireline
(deployed directly in the wellbore) located in the Codell well C1 highlighted by the green pentagon, and (3)
7
|
Colorado School of Mines
|
Figure 2.2 Gunbarrel view of the well configuration: zipper one wells (green) and zipper two wells (white).
Fiber installations include permanent (N1B), temporary wireline (C1), and temporary disposable (N3C).
a temporary single-use fiber (deployed directly in the wellbore) located in the Codell well C3 highlighted by
the pink pentagon. The permanent fiber well N1B was treated during zipper group one, acquiring in-well
DAS and DTS data during hydraulic fracturing stages. The temporary wireline fiber well C1 and
temporary single-use fiber well N3C monitored the zipper one treatments, acquiring far-field low-frequency
distributed acoustic sensing (LF-DAS) signals induced by the nearby injection wells. Additional completion
data such as pumping curves (i.e., slurry rate, proppant concentrations, etc.) and high-frequency (10 Hz)
pressure gauge data (treatment-pressure) were also acquired.
This study will focus on the subset of stages 17-20 during the treatment of well N1B over the course of
approximately 24 hours. Three sets of DFOS data are available for each of the stages: in-well DAS and
DTS acquired from N1B in the Niobrara B, and LF-DAS cross-well strain acquired from the temporary
fiber well C1 in the Codell. The perforation designs were common to all stages. Cluster spacing was
approximately 23 ft, with 12 evenly spaced clusters across 300 ft compartment lengths. Approximately 750
lbs/ft proppant was injected for stages 17-18, and 1000 lbs/ft for stages 19-20, at a maximum rate of 50
bpm.
2.3 In-well DTS Fiber-optic Monitoring
Continuous DTS data were acquired during completion providing the dynamic temperature profile
along the Niobrara B N1B wellbore. The DTS profile during the hydraulic fracturing treatment of stages
8
|
Colorado School of Mines
|
17-20 are shown in Figure 2.3. Time is along the x-axis and measured depth along the y-axis, increasing
from top to bottom (heal to toe). Each stage can be identified in depth and time by the red dashed line,
indicating the plug set depth and injection period. The amplitudes in Figure 2.3 indicate the temperature
spatially along the fiber over time, with red for the warmest, and dark blue as much cooler.
The cable was installed behind the wellbore casing, allowing for DTS to be recorded during the
injection cool-down and post-treatment warm-back period [26]. During hydraulic fracturing treatments,
DTS measures the cooling along the entire wellbore caused by the injection of relatively colder fluids and
proppants. During a frac-stage, fluid and proppant pass through the heel side causing near uniform cooling.
The coolest temperatures are recorded near the stimulated perforation clusters, indicating more direct fluid
contact with the optical fiber cable [21, 25]. Post-treatment thermal recovery (warm-back) is observed after
injection stops and the near-wellbore region returns to the geothermal surrounding temperature.
Figure 2.3 Dynamic temperature profile acquired by in-well DTS installation during the hydraulic
fracturing treatment of stages 17-20 in well N1B.
We use the indicated bridge plug depths to identify the targeted stage interval. This provides quick,
qualitative information describing interstage fluid communication. We then calculate the depth-averaged
temperature for each stage during each hydraulic fracturing treatment period. The result is a temperature
curve for the current and subsequent treatment stages. The difference between the temperature at the
start and end of each treatment period serves as a static, qualitative attribute indicating the fluid intake at
each stage interval.
9
|
Colorado School of Mines
|
2.4 In-well DAS Fiber-optic Monitoring
Figure 2.4, in which acoustic energy is plotted against time, shows the in-well DAS data acquired
during the treatment of stages 17-20. The axes are common with the DTS plot (Figure 2.3), with the plug
set depths indicated by the black dashed lines. The red/white in Figure 2.4 represent the high acoustic
intensity amplitudes recorded along the DAS fiber at each perforation cluster. The use of in-well DAS for
hydraulic treatment diagnosis relies on the correlation of fluid flow rate through perforation clusters and
acoustic signatures [21, 24].
Figure 2.4 In- well DAS profile acquired during the hydraulic fracturing treatment of stages 17-20 in well
N1B.
Processing DAS data into different frequency bands is common practice as it allows for the
investigation of different physical concepts in and near the wellbore [18]. We utilize the 500-5000 Hz
frequency band processed and delivered by the service provider for stage isolation analysis. At this
frequency band, acoustic energy amplitudes are assumed to be dominated by “perforation entry noise” and
correlate to perforations taking fluid and proppant during the injection period [18, 21]. We identify the
bridge plug depths to determine the targeted treatment interval.
We then calculate the RMS amplitudes spatially along the fiber, summing over the treatment period
according to:
(cid:118)
(cid:117) N
(cid:117)1 (cid:88)
X =(cid:116) x2 , (2.1)
RMS n i,j
i=1
10
|
Colorado School of Mines
|
Figure 2.5 Processed low-frequency strain data (top) recorded at offset well C1 during the completion of
stages 16-20 in treatment well N1B. The slurry rate (bottom) is plotted for all zipper one wells.
2.6 Strain Inversion
We applied a geomechanical inversion algorithm presented by Liu et al., (2021a, 2021b) to constrain the
dynamic fracture widths using LF-DAS data recorded at the offset well C1. While the reader is referred to
Liu et al. (2021a, 2021b) for a detailed formulation, the major concepts are summarized below:
• The displacement discontinuity method (DDM) is used to construct a forward model relating
LF-DAS strain data to hydraulic fracture geometry. DDM efficiently calculates fracture induced rock
deformations in an elastic body.
• The fracture/fiber system is discretized into N fracture elements and M sensing points.
• Assuming linear elastic rock deformation, the strain recorded at a sensing point M, is the
superposition of strain contributions from N fracture elements.
• Linear least-squares inversion is used to solve the system of equations for fracture width at discrete
timesteps. The LF-DAS data, stored by optical-phase change rate, was converted to strain rate prior
to inversion as follows [29]:
λ
(cid:15). = (cid:52)φ, (2.2)
4πnξL
where n=1.5 the refractive index, dimensionless; ξ =0.8 a multiplicative constant, dimensionless; L=5 m
the gauge length; and λ=1550 nm which is the probe wavelength.
12
|
Colorado School of Mines
|
Several input parameters must be considered prior to inversion. A parallel orientation between the
injector well N1B and monitor well C1 were common in all stages. A 16.4 ft gauge length was used for the
LF-DAS. Horizontal and vertical offset were approximately 90 ft and 210 ft, respectively. However, the
horizontal and vertical offset values did vary slightly based on the treatment stage. The fracture half-height
and half-length were both estimated to be 300 ft. The half-height must be at minimum the vertical
distance between wells N1B and C1, likely propagating past C1 given the signal strength observed in
LF-DAS Figure 2.5. The fracture half-length was constrained using the horizontal offset (approximately
280 ft) between well N1C and N1B as a proxy. Frac hit signals were observed in LF-DAS Figure 2.5
consistently during the treatment of well N1C. A Poison’s ratio of ν =0.29 was used based on previous
work in the nearby Chalk Bluff field as part of RCP’s Phase XVIII field project.
We then determined individual fracture hits for treatment stages. Figure 2.6a shows the LF-DAS
recorded from monitor well C1 during treatment stage 18 as an example. The dashed lines represent the
picked fracture hits as follows: (1) closing fractures from stage 16 (black); (2) reactivated fractures from
stage 17 (blue); (3) new fracture hits from stage 18 (yellow); and (4) fracture hits from the Codell well C3
stimulated at the same time (purple). The process for picking fracture hits in a complex dataset rely on:
• The extension zone at the tip of the fractures were frequently observed as a heart-shaped extending
pattern at the beginning of the fracture hit signal [4, 14].
• The observed polarity reversal during the step-down, and at the end of fluid injection represent
fracture opening (red amplitudes) and closing (blue amplitudes) [4, 14].
• Previous stage fractures are indicated by either: (1) closing signature for the duration of the injection
period; or (2) closing signature before injection, and opening signature after injection begins
(reactivated fractures).
• Fracture hits and pumping curves were compared temporally to distinguish fracture hits from N1B
and C3.
It is worth noting that the fracture hits from well C3 were included during inversion processing, but the
results were not considered for this study. This is because (1) there was no in-well data (DAS/DTS) for
C3, and (2) the inversion parameters differ from well N1B making the inverted widths unreliable.
The strain rate (Figure 2.6a) was then converted to strain change (Figure 2.6b) by integrating in time.
The strain data near the fracture hit locations were removed because decoupling and thermal effects in
that region can make the inversion results unreliable [13]. The model predicted strain change (Figure 2.6c)
was then calculated using the inverted widths to ensure agreement between the field and modeled data.
13
|
Colorado School of Mines
|
Figure 2.6 (a) Strain rate waterfall plot, including picked frac hits for stage 16 (black), stage 17 (blue),
stage 18 (yellow), and well C3 (magenta). (b) strain change waterfall plot. (c) Calculated strain change
waterfall plot using inverted widths.
The result is a dynamic width profile for each picked fracture during the injection period. To better
diagnose stage isolation, we grouped together fractures common to each stage and summed the widths at
each time step. This provides dynamic width profiles at a stage level as opposed to a fracture level. It
should be noted that the spatial resolution (5 m) of DAS make it difficult to distinguish if each fracture hit
is caused by a single fracture or multiple fractures [4]. In this study, we will assume that each fracture hit
is from a single fracture. Because we grouped fractures by stage, the effect on the results is negligible.
2.7 Results
2.7.1 Treatment Stage 17
Figure 2.7 shows the results for stage 17. The top track displays the LF-DAS (Figure 2.7a) and in-well
DAS (Figure 2.7b), the middle track displays the width inversion results (Figure 2.7c) and in-well DTS
(Figure 2.7d), the bottom track displays the pumping curve information (pressure, slurry, and proppant
concentrations) in relative values (Figure 2.7e-f). The in-well DAS, in-well DTS, and LF-DAS plots share
the same y-axis (12,000 – 14,500 ft). It should be noted that there is a break in the optical fiber (∼14,250
ft) that causes a gap in the DTS data (Figure 6d) and high amplitude noise in the in-well DAS (Figure 6b).
All plots share the same x-axis (clock-time). The black and red dashed lines indicate the plug set depth of
the current hydraulic fracturing stage.
The DTS profile (Figure 2.7d) indicates good isolation of the current stage from the previous frac stage
as no fluid induced cooling (dark-blue) goes beyond the plug set depth. This is further supported by the
gradual warming (green to yellow) toe ward from the plug depth. Additionally, the near uniform cooling
within the targeted interval indicates that all perforation clusters took in fluid.
14
|
Colorado School of Mines
|
Figure 2.7 Results for stage 17, including: (a) LF-DAS waterfall plot; (b) in-well DAS waterfall plot; (c)
inverted fracture widths grouped by stage interval; (d) in-well DTS waterfall plot; (e)-(f) pumping curves.
The acoustic intensity recorded by DAS (Figure 2.7b) during the step-down rate tests and main frac
treatment indicate that all 12 clusters took in fluid to some degree. At approximately 1.5 hours the
treatment pressure drops coincident with the bottom-hole (BH) proppant concentration increase.
Treatment pressure gradually declines while BH proppant concentration incrementally increases for the
duration of the treatment. Approximately 2 hours into the injection period, acoustic intensity fades rapidly
across all perforation clusters, with only three or four clusters taking most of the fracking fluid by the end
of the treatment. Favorable stage isolation is indicated by the lack of acoustic intensity signal below the
plug set depth, agreeing with the DTS results.
Newly stimulated fractures within stage 17, and previous fractures from stages 15 and 16 are observed
in the LF-DAS (Figure 2.7a). The high amplitude extending strain signal indicates the fractures in stage
17 were opening during the frac treatment. In contrast, the strong compressing signal in stage 16 indicates
fracture closure. The extending/compressing strain signals heel ward from stage 17 are determined to be
fractures propagating from the treatment of Codell well C3.
15
|
Colorado School of Mines
|
The dynamic fracture width profile results are as follows: (1) newly stimulated fractures in stage 17
grow to a sum width of 2.26 mm; (2) fractures from stage 16 close by 1.02 mm; and (3) negligible response
is recorded from stage 15 fractures. During the hydraulic fracturing stage, the following observations are
noted:
• The inverted fracture width for stage 17 (blue line) indicates strong sensitivity to the in-well injection
rate highlighted by the step-down test. The width increase ceases when treatment pressure drops,
and increases again when injection resumes.
• Negative width values for stage 16 (black line) indicate fracture closure during the hydraulic
fracturing stage (see discussion section).
2.7.2 Treatment Stage 18
Figure 2.8 shows the results for stage 18. Note that the red square along the y-axis of the in-well DAS
and DTS (Figure 2.8b and Figure 2.8d) indicate the bridge plug depth of the preceding stage. The in-well
DTS profile (Figure 2.8d) shows uniform cooling in the stage 18 interval, suggesting fluid distribution
across the stage interval.
The cooling trend toe-ward past the plug depth indicates injection fluid flow into the previously
fractured stage. Uniform cooling occurs approximately 20 minutes into the injection period at the stage 18
depth interval while stage 17 cooling occurs gradually toe-ward later into the treatment. Acoustic intensity
recorded by DAS (Figure 2.8b) indicate that all 12 clusters took fluid at the start of injection, with most of
the fluid flow through the toe-side clusters nearest the plug. Immediately before and after the step-down
test, several perforation clusters in the previous stage are reactivated. Approximately 1.5 hours into the
injection, there is a sharp decrease in treatment pressure (∼1000 psi) coincident with acoustic intensity
fading in stage 18 clusters and an increase in acoustic intensity in previous stage clusters. Acoustic
intensity diminishes sharply 2.0 hours into the treatment, and it becomes unclear which clusters were
effectively stimulated.
The LF-DAS data (Figure 2.8a) show extending strain at stage 17 and 18 depths and compressing
strain at stage 16 depths; indicating opening and closing fractures, respectively. The dynamic width
profiles in Figure 2.8c indicate: (1) newly stimulated fractures in stage 18 grow to a width of 2.0 mm; (2)
reactivated fractures in stage 17 increase width by 1.2 mm; and (3) fracture widths in stage 16
continuously decreased by 2.1 mm. An inflection in the width profile of stage 17 indicates an increase in
the growth rate coincident with the treatment pressure drop. This trend is also observed in the LF-DAS
data as a sudden increase in the extending strain at stage 17 depths (Figure 2.8a).
16
|
Colorado School of Mines
|
Figure 2.8 Results for stage 18, including: (a) LF-DAS waterfall plot; (b) in-well DAS waterfall plot; (c)
inverted fracture widths grouped by stage interval; (d) in-well DTS waterfall plot; (e)-(f) pumping curves.
2.7.3 Treatment Stage 19
Figure 2.9 shows the results during the treatment of stage 19. The in-well DTS profile (Figure 2.9c)
shows uniform cooling in the stage 19 interval 30 minutes after injection begins. Fluid continues toe-ward
beyond the plug depth into stage 18, uniformly cooling the interval. Fluid also continues into stage 17,
where cooling is observed at only half of the interval, suggesting heel-side clusters took more fluid.
Acoustic intensity recorded by DAS (Figure 2.9b) indicate all clusters were stimulated pre and post
step-down. After 1.0 hour of injection the acoustic amplitude abruptly diminishes coincident with a sharp
drop in treatment pressure (∼1700 psi). Intensity gradually fades during injection, with the heel-side
cluster(s) taking most of the fluid as indicated by the high amplitudes near the plug depth. Qualitative
observations suggest DAS does not agree with the DTS profile, as there is no indication that perforation
clusters were stimulated in the previous stages during injection.
17
|
Colorado School of Mines
|
Figure 2.9 Results for stage 19, including: (a) LF-DAS waterfall plot; (b) in-well DAS waterfall plot; (c)
inverted fracture widths grouped by stage interval; (d) in-well DTS waterfall plot; (e)-(f) pumping curves.
LF-DAS (Figure 2.9a) indicates new fracture openings in stage 19, reactivated fractures in stages 17
and 18, and fractures closing in stage 16. The dynamic width profile results are as follows: (1) newly
stimulated fractures in stage 19 grow to a fracture width of 0.8 mm; (2) reactivated fractures in stage 18
increase width by 0.9 mm; (3) stage 17 fractures decrease width by 0.3 mm; and (4) stage 16 fractures
decrease width by 0.2 mm. Several complex fracture interactions are observed in the dynamic width
response and LF-DAS during the injection period:
• An inflection point occurs in stages 17-19 coincident with the treatment pressure drop at 1.0 hours
into injection. Stage 19 fracture width growth rate decreases while stages 17 and 18 width growth
rate increases. Stage 17 notably changes from decreasing width to an increasing width trend.
• Qualitative LF-DAS observations indicate increased extending strain in stages 17 and 18 coincident
with the treatment pressure drop.
18
|
Colorado School of Mines
|
• A noticeable increase in treatment pressure at 1.5 hours into the injection are coincident with a
decreased width growth rate in stage 19, another increase in width growth rate in stage 18, and stage
17 changes from an increasing to decreasing width trend.
2.7.4 Treatment Stage 20
Figure 2.10 shows the results during the injection of stage 20. Note that stage 20 treatment was
sectioned into two pumping schedules (20a and 20b). The DTS profile (Figure 2.10d) shows uniform
cooling in the stage 20 depth interval. The cooling trend continues toe-ward past the plug depth extending
through stage 18. Stage 17 appears to take in fluid at the most heel-ward perforation cluster midway
through injection period 20a , with the rest of the stage interval showing a warming trend.
Figure 2.10 Results for stage 20, including: (a) LF-DAS waterfall plot; (b) in-well DAS waterfall plot; (c)
inverted fracture widths grouped by stage interval; (d) in-well DTS waterfall plot; (e)-(f) pumping curves.
Acoustic intensity recorded by DAS (Figure 2.10b) indicate all clusters were stimulated pre and post
step-down with the toe-ward cluster nearest the plug intaking the most injection fluid. Acoustic intensity
19
|
Colorado School of Mines
|
gradually fades uniformly across clusters during injection period 20a and is significantly diminished during
injection period 20b . Low amplitude acoustic intensity indicates restimulated perforations in stage 19
interval during treatment period 20a. During treatment 20b, there is an abrupt perf reactivation in stage
18 at clock time 19:30 but diminishes rapidly.
The LF-DAS (Figure 2.10a) indicates newly stimulated fractures in stage 20, and restimulated fractures
through stage 17. The dynamic width results are as follows: (1) stage 20 fracture width grows to 0.6 mm;
(2) stage 19 fractures are reactivated, growing 0.25 mm, (3) stage 18 fractures are reactivated, growing 0.5
mm; and (4) stage 16 and 17 fractures follow a closing trend, reducing 0.2 and 0.4 mm, respectively.
Several observations from the dynamic width profiles worth noting are:
• At 1.5 hours into treatment 20a, stage 17 width changes from decreasing to increasing, which agrees
with DTS and LF-DAS profile.
• The beginning of treatment 20b results in the fracture width increasing in stage 18, followed by stage
19, and finally stage 20. This response sequence supports far-field strain as a function of in-well fluid
distribution.
• The fracture growth in stages 18-20 increases the closure rate of stage 16 and 17 indicated by the
width results following the start of treatment 20b.
• Signals from Codell well C3 are visible in the LF-DAS during treatment 20b.
2.8 Discussion
Extendingouranalysisacrossmultipletreatmentstagespermitstheintegrationofin-wellandoffset-well
methods for a comprehensive diagnosis. Our study indicates robust agreement between in-well DTS and
LF-DAS inversion results. Figure 2.11a shows the cumulative fracture width changes for each substage,
recorded at the end of the targeted treatment stage. Positive and negative column values represent fracture
opening and fracture closure, respectively. Similarly, Figure 2.11b represents the spatially averaged
temperature difference between the start and end of the targeted treatment stage. Positive column values
represent cooling from fluid contact, and negative values represent warming from favorable stage isolation.
DTS and LF-DAS inversion indicates in three of four stages monitored, fluid leaked below the plug, and
stage isolation was not complete. During the hydraulic fracturing treatment of stages 18-20, fluid flows into
one or more of the previous stage compartments, cooling that section of fiber to near fluid temperatures.
Accordingly, the far-field response is constrained by LF-DAS inversion, showing a decrease in stimulated
fracture widths across the actual treatment stages. During the treatment of stage 17, isolation is complete,
and no fluid flow past the bridge plug is observed. Showing strong agreement, LF-DAS inversion indicates
20
|
Colorado School of Mines
|
optimal fracture growth in the targeted treatment stage, while stage 16 fracture widths decreased
significantly. However, during the treatment of stages 18-20, inter-stage fluid communication extends into
adjacent stage intervals, and fracture widths for both the targeted and previous treatment stages diminish.
In general, we observe a 73% difference in fracture width growth for the targeted treatment stage between
the stages 17-20.
The observed decrease in fracture width is likely due to near-wellbore region (NWR) and far-field
effects. In the NWR, inter-stage fluid communication effectively increase the compartment length from 300
ft to 900 ft, affecting cluster efficiency, fluid distribution and proppant placement [30]. In the far-field,
opening fractures suppress the nearby rock formation, leading to the closure of nearby fractures [12].
In contrast, assuming a relationship between fluid flow rate and acoustic intensity, in-well DAS shows
little evidence of poor stage isolation. Figure 2.11c shows the normalized acoustic energy summation for all
current and previous hydraulic fracturing stages, including the calculated noise floor for each.
Theoretically, high acoustic intensity at perforations below the stage plug of the targeted interval describe
the degree of fluid flow into previous stages [22, 24]. While the acoustic intensity follows the same
diminishing trend for the targeted treatment interval (∼35% between stage 17 and 20), there is little
indication of cluster restimulation. Interpretations based on in-well DAS alone would suggest that isolation
issues occur only during the treatment of stage 18 (Figure 2.11c, Figure 2.8b).
The absence of DAS intensity at restimulated clusters is likely due to variations in the wellbore region.
The acoustic intensity fading in our results (Figure 2.7b, Figure 2.8b, Figure 2.9b, Figure 2.10b) is
commonly observed when proppant reaches the perforation clusters [26]. Proppant induced erosions occurs,
beveling and smoothing the edges of the perforation holes, reducing the amount of friction and noise
despite treating pressure remaining relatively constant [26, 31]. Accordingly, the acoustic intensity
generated by reactivated clusters is very weak and can hardly be differentiated from the background noise
recorded by DAS.
The sensitivity of the LF-DAS inversion to in-well variations is highlighted by stages 18 and 19
(Figure 2.8 and Figure 2.9). Midway through hydraulic treatment, a significant pressure drop is observed
in the pumping curves coincident with an inflection point in the LF-DAS inversion results. Specifically, we
see previous stage fracture widths grow at a faster rate than the targeted treatment stage. The response is
also observed in LF-DAS (Figure 2.8a and Figure 2.9a) as an abrupt extension signal. This indicates that
the pressure drop occurs due to inter-stage fluid communication, the effects of which are observed in the
far-field and quantitatively constrained by restimulated fracture widths. Important to note is that the
in-well DAS (Figure 2.8b and Figure 2.9b) show a significant drop in acoustic intensity at the targeted
clusters but no noticeable change below the bridge plug depths. This could be misleading, suggesting that
22
|
Colorado School of Mines
|
the pressure drop is due to an increase in BH proppant concentraions and/or proppant induced erosion of
the perforation holes.
The implications of our findings demonstrate LF-DAS recorded in an offset well is a viable method for
multi-stage hydraulic fracturing treatment diagnosis. LF-DAS inversion can be used to constrain far-field
fracture widths to evaluate completion efficiency and design optimization. In-well variations (i.e.,
inter-stage fluid communication, bridge plug failure) can be identified and quantitatively described in the
far-field through fracture width inversion. Volumetric estimates can be made by placing additional
constraints on restimulated fractures (i.e., fracture shape) to determine the fluid distribution between
current and previous stage fractures.
Installation of permanent in-well optical-fiber is expensive and requires careful consideration of
placement along the wellbore to avoid damage during hydraulic fracturing [18]. Temporary wireline
deployments in an offset well are considerably less expensive and require less considerations during
installation (i.e., cementation to formation, configuration). In operations where in-well DAS is deemed too
costly or there is adverse risk, temporary fiber installations can be used to monitor hydraulic fracture
treatments.
2.9 Conclusions
We demonstrate the effectiveness of using DAS installed in an offset well to diagnose multi-stage
hydraulic fracturing treatments. We apply a geomechanical based inversion algorithm to the low-frequency
strain data to quantitatively describe the time evolution of fracture widths in the far-field. We use the
inverted fracture widths to diagnose stage isolation and bridge plug failure and verify our results using
in-well DTS and DAS. Our major findings include:
• In-well DTS and cross-well LF-DAS inversion show strong agreement regarding plug integrity
evaluation, indicating poor stage isolation and/or bridge plug failure in three of four investigated
treatment stages.
• Poor stage isolation results in significantly decreased fracture widths recorded at the monitor well in
the current treatment stage. This is due to decreased in-well cluster efficiency, and far-field stress
shadowing.
• In-well DAS fails to diagnose inter-stage fluid communication in all but one of the treatment stages.
The absence of reactivated perforation noise is likely due to proppant induced erosion effects during
treatment.
23
|
Colorado School of Mines
|
CHAPTER 3
RAPID SURFACE-DEPLOYMENT OF A DAS SYSTEM FOR EARTHQUAKE HAZARD
ASSESSMENT
3.1 Introduction
Variations in near-surface soft soil deposits are thought to largely dictate seismic amplitudes on a local
scale [2, 32, 33]. The time-averaged shear-wave velocity of the top 30 m (VS30) has been the quantitative
parameter adopted by NEHRP (National Earthquake Hazards Reduction Program) for site classification
and building codes in the United States for nearly two decades [34–36]. Site class is used to characterize soil
properties and determine the local seismic coefficients for earthquake-resistant structural design [35, 37].
For large-scale commercial buildings and city infrastructure, the difference between site classifications can
substantially affect the initial and long-term cost of construction [37]. VS30 has also been widely used as a
parameter for characterizing effects of soil stiffness on ground-motion prediction studies [38–40].
Traditional VS30 surveys generally are invasive, use logistically challenging borehole measurements (i.e.,
crosshole, suspension logging methods), and require prior authorization to implement [35, 36, 41].
Accordingly, current VS30 maps are spatially limited in coverage. In areas where no VS30 data are
available, estimates often rely on interpolated values or broadband and long-period permanent seismic
arrays that suffer from reliability and resolution due to poor station distribution [2]. The development of
non-invasive techniques that measure surface wave dispersion properties address many of the challenges
associated with VS30 surveys and greatly improve data coverage and availability for structural engineers.
Within the last two decades, the spectral-analysis-of-surface-waves (SASW) method has been widely used
for cost-effective, non-invasive site classification studies [37, 41–43].
In recent years, distributed acoustic sensing (DAS) has emerged as a reliable tool for near-surface
applications and subsurface characterization. Its versatility and high spatial-temporal resolution make it a
suitable instrument for non-invasive VS30 surveys. DAS uses the principles of time-domain reflectometry
to effectively turn a length of a fiber-optic cable into a linear network of seismic sensors out to tens of
kilometers [2]. Dynamic strain is recorded along the fiber by measuring the phase difference of
backscattered light caused by impurities in the fiber core. DAS is mostly sensitive to particle motion along
the axis of the fiber, making it highly effective for measuring the horizontal component of surface waves
(i.e., Rayleigh). Recent applications use pre-existing unused telecommunication networks (i.e., so-called
“dark fiber”) to record high-quality seismic data from earthquakes [7–9] as well as ambient and
anthropogenic sources in dense urban areas [44–46]. In locations where dark fiber is unavailable, traditional
25
|
Colorado School of Mines
|
deployment methods consisting of burying (<1 m) DAS cables [1, 17] are used to ensure sufficient coupling
and higher signal-to-noise ratios. However, this approach often proves cost prohibitive and logistically
challenging to implement.
Research concerning the instrument response of DAS cables deployed directly on the ground surface is
sparse. Spikes et al. 2019 demonstrate that helically wrapped cables draped along the ground surface
recorded similar quality active-source (i.e., hammer shots) reflection data as collocated geophones.
High-fidelitydataacquisitionfromsurfaceDASarrayscouldprovidegeotechnicalengineerswithanefficient
and cost-effective method for numerous near-surface applications. In this study, we compare ambient
waveform data recorded by a surface-deployed DAS array with those recorded from collocated trenched
cables. We adopt a simplified SASW method to estimate VS30 for linear subsections of both arrays.
The paper begins with a description of the unique deployment of the DAS arrays and the data
acquisition parameters used for the field study. We then discuss our data processing approach, which
includes ambient-data interferometry, dispersion curve mapping, and generating VS30 estimates. This is
followed by a presentation and discussion of our results for each data processing step. We find that further
analysis of the ambient waveforms and dispersion curves provides important qualitative information such
as signal coherency and dispersion features to compare the different coupling conditions. A quantitative
comparison of our VS30 results shows that using surface-deployed DAS cables is a viable approach for
near-surface geotechnical surveys. We conclude with potential implications for the application of surface
DAS fiber deployments.
3.2 Field Description and Data Acquisition
The field test was conducted at Kafadar Commons located on the Colorado School of Mines campus.
Kafadar Commons is a rectangular grass field approximately 40 m by 100 m in size. At the time of data
acquisition, the southern half of the field was covered by several inches of snow while the northern side was
exposed grass.
We deployed 650 m of optical fiber along the surface of the field and connected it to the 1-km trenched
DAS array previously installed beneath Kafadar Commons as part of the Mines Underground Geophysical
Laboratory (see Figure 3.1) [47]. The surface array was deployed as six parallel lines roughly 100 m in
length using two different coupling methods: (1) laid directly on the ground; and (2) pressed into the
ground by several people walking on top of the cable. Cables deployed under condition (2) will be referred
to as ‘Pressed’. The first line segment was draped along the cement walking path north of the field
(Cement North), followed by two lines along the grass (Grass Pressed and Grass), and two lines along the
snow (Snow and Snow Pressed). The final line segment was deployed along the cement walking path south
26
|
Colorado School of Mines
|
Figure 3.1 DAS deployment overview. (a) Map view of Kafadar Commons showing DAS surface and
subsurface deployments. Photographs showing the different surface deployments with black lines
superimposed on cables, including (b) Cement north, (c) Grass (left) and Grass Pressed (right), and (d)
Cement South, Snow, and Snow Pressed (left to right).
of the field (Cement South) and terminates inside the adjacent CoorsTek building (see Figure 3.1a).
Parallel line segments were separated by approximately 2.5 m, except the Cement North (Figure 3.1b) and
Grass Pressed (Figure 3.1c) segments which were separated by approximately 7.5 m, and the Grass
(Figure 3.1c) and Snow Pressed segments by 30 m (Figure 3.1d). The overall footprint of the surface array
is 45 m by 100 m.
The trenched array follows a 27 m by 90 m grid pattern located 1 m beneath the surface (Luo et al.,
2020). To examine nearly coincident instrument responses, we only use data recorded along the four
subsections (Subsurface 1-4) running parallel to the surface array. The two subarrays were connected by
splicing the end of the surface cable to the fiber-optic lead of the subsurface cable located in the CoorsTek
building. We calibrated the DAS channel locations using GPS coordinates, tap-tests, and hammer-shot
data.
The range of detectable wavelengths is constrained by the receiver (DAS channel) spacing ∆x and
min
the array aperture ∆x along the linear subsections of the array. Wavelengths shorter than twice the
max
channel spacing of the array will be spatially aliased according to the Nyquist theorem. In general, the
longest resolvable wavelengths are twice the array aperture. For the frequency-phase velocity relationship,
27
|
Colorado School of Mines
|
the aliasing and resolution boundaries can be expressed as 2f ∗∆x and 2f ∗∆x ([1]).
min max
Approximately 1.0 hour of data were recorded on December 2nd, 2020, using a Terra15 Treble
interrogator unit at a 4467 Hz sampling rate with a channel spacing of approximately 2.45 m. A gauge
length of 4.0 times the channel spacing was used, which acts as a differential operator across the spatial
axis of the data to convert the measurements to strain rate equivalent. We recorded ambient waveforms
from anthropogenic sources in the survey area, including a nearby road running orthogonal to our array
setup approximately 5 m from the nearest DAS channel.
3.3 Methodology
3.3.1 Ambient-noise Interferometry
We use ambient seismic interferometry [48, 49] to produce virtual shot gathers for each linear profile of
the surface and subsurface arrays. With this approach, each DAS channel along a linear segment of fiber is
considered a virtual seismic source (U ) or receiver (V ). We partition the 1-hr record of data into 2.0-s
i i
time windows and then calculate and stack the cross-spectrum [50] according to
ρ(x ,x
,ω)=(cid:88)N U i(x s,ω)V i(cid:63)(x r,ω)
(3.1)
s r |U (x ,ω)||V (x ,ω)|+(cid:15)
i s i r
i=1
where ω is angular frequency, U and V are the Fourier transform of the ith time segment of two virtual
i i
stations, and x and x are the spatial locations of the virtual source and receivers, respectively. (cid:63) denotes
s r
the complex conjugate, and real positive constant (cid:15) is a stabilization term. Individual cross-spectra are
calculated for all possible station pairs, resulting in 7162 virtual shot-receiver pairs. The 2.0-s windows are
stacked in the frequency domain, with the resulting cross-spectrum returned to the time domain by
applying an inverse Fourier transform. The resulting waveform has positive and negative time-lags
representing waves propagating in opposite directions between the virtual stations [49]. Applying this
method within each linear subsection and treating each DAS channel as a virtual source produces
corresponding 3-D waveform volumes that are analyzed for data quality prior to dispersion analysis.
3.3.2 Dispersion Analysis
To produce frequency-velocity dispersion images we use the phase-shift method [48]. With this method,
we can separate the different surface-wave modes despite the relatively limited number of traces and offsets
from our virtual shot gathers. The first step is to extract the wavefield phase by applying a temporal
Fourier transform to the virtual shot gathers (ρ) produced in equation 3.1. Next, we apply the
offset-dependent phase shift and normalize the amplitudes to ensure equal weighting for each individual
28
|
Colorado School of Mines
|
trace:
P(ω,c)=(cid:88)(cid:88) ρ(x s,h,ω)
e−iωh/c, (3.2)
|ρ(x ,h,ω)|
s
xs h
where h=x −x represents a vector of station offsets. The maximum amplitude term occurs when signals
s r
travel at the same phase velocity (c) for a given angular frequency. This produces a frequency-velocity
dispersion image for each channel serving as a virtual source along each linear profile. Because the primary
goal is to produce VS30 estimates for each linear subsection, we stack the dispersion images of all virtual
sources from the same linear profile to produce a single dispersion image for each subsection.
3.3.3 Estimating VS30
We estimate VS30 for each subsection using a simplified SASW method [34]. Compared to the
traditional SASW approach, this method is computationally efficient but provides only a single VS30 value
without using traditional inversion methods that produce 1D shear-wave velocity profiles for a range of
depths. The correlation between Rayleigh-wave phase velocity and VS30 can be described by the predictive
empirical equation [34]:
VS30=1.076×VR36, (3.3)
where VR36 is the Rayleigh wave phase-velocity at a wavelength λ=36 m. Extracting VR36 from the
processed dispersion curves results in a phase velocity and amplitude for discrete frequencies within the
chosen frequency band. Frequency resolution is determined by two parameters: (1) the raw data sampling
rate and as consequence of the Fourier transform, and (2) the window length of the processed ambient time
series. A cubic interpolation method was applied to each dispersion image to improve the frequency
resolution prior to VS30 processing. We determine the final VS30 value for each subsection using the VR36
value with the highest amplitude in the dispersion image.
3.4 Results
Figure 3.2 displays the examples of ambient waveforms for surface array subsections (a)-(f) and two
subsurface array subsections (g)-(h) after transforming the cross-spectra to the time-space domain. Offsets
for each subsection vary slightly due to differences in cable length. The observed waveform direction
reversal is due to subsections being deployed in an east-west or west-east orientation. The high-amplitude
zero offsets of the waveforms represent the source channel autocorrelation of each virtual shot gather. After
applying a 2-20 Hz bandpass filter and manually scanning for data quality, we find most virtual shot
gathers exhibit coherent waveforms. Figure 3.2 highlights offsets with poor signal response. Notably, the
Snow subsection lacks coherent waveform features along 40 m of the fiber and, to a lesser extent, the Grass
29
|
Colorado School of Mines
|
Pressed and Snow Pressed subsections display poor data quality along 12 m of fiber. The waveforms for all
subsections display an asymmetric distribution, which is discussed in detail below.
Figure 3.3 shows the stacked, normalized dispersion curve results for the surface (a)-(f) and subsurface
(g)-(h) profiles for the 3-20 Hz frequency band and a 100-2000 m/s phase velocity range. The most stable
results are within the lower frequency range of 3-7 Hz defined by the high spectral amplitudes in red. All
dispersion curves show an abrupt discontinuity between 7-9 Hz, potentially indicating a mode transition
[1]. Coherency in the dispersion curves begins to diminish at higher frequencies, with no continuous
dispersion features above 20 Hz. The subsurface subsections show the most consistent results with
relatively well-defined spectral amplitudes for frequencies below 15 Hz. These are followed by the Cement
North, Grass, and Grass Pressed subsections that show stable results for frequencies less than 12.5 Hz. The
Snow Pressed subsection is stable for frequencies less than 11 Hz, and finally the Snow and Cement South
subsections are stable for frequencies below 10 Hz.
The Rayleigh wave phase velocity primarily depends on the subsurface geomechanical properties (i.e.,
shear-wave velocity, and to a lesser extent, compression-wave velocity, and density) down to a depth of one
wavelength, with maximum sensitivity at approximately 1 to 1 the wavelength ([34]). Accordingly, shorter
2 3
wavelengths describe the shallower subsurface, with longer wavelengths providing material properties at
greater depths. The frequency-phase velocity maps (Figure 3.3(a)-(f)) for the surface subsections provide
reliable estimates between wavelengths of approximately 40 m and 125 m (4 Hz - 10 Hz). In some sections,
frequencies beyond 10 Hz become less reliable due to the lack of coherent dispersion features. The
subsurface sections (Figure 3.3(g)-(h)) demonstrate coherent dispersion features up to 15 Hz corresponding
to a wavelength of approximately 20 m. The maximum wavelength that can be recovered from seismic
interferometry is limited to the array size, due to the interference between forward and backward
propagating wavefields. In this study, however, we are able to make reliable measurements to the
wavelengths longer than the linear sections. One potential reason is that because the noise sources are
biased to one-side of the array, the effect of interference is minimized.
VR36 values were estimated using the dispersion curves for each subsection. Figure 3.4 shows the
results for the Cement North profile as an example. The dashed line in Figure 3.4a represents the
phase-velocity and frequency values corresponding to a wavelength of 36 m (VR36). Figure 3.4b tracks the
normalized spectral amplitude as a function of phase velocity for discrete frequencies along the VR36 line.
The marker denotes the peak amplitude which is the value used to calculate VS30 for each subsection
(Figure 3.5). Note that Grass Pressed and Cement South returned peak spectral values along the VR36
line corresponding to a phase velocity of 279 m/s and 705 m/s, respectively. Prior to calculating the VS30
value for these subsections, we analyzed their dispersion curves and found that the second highest spectral
30
|
Colorado School of Mines
|
values compared well to the phase velocities of adjacent subsections, which are plotted in Figure 3.5
Figure 3.4 Estimated VR36 results for the Cement North surface subsection. (a) shows the stacked
normalized dispersion curve. The dashed line indicates λ=36 m for corresponding phase-velocities and
frequencies. (b) indicates the spectral amplitudes for frequency values along the dashed line from (a) with
the marker indicating the peak λ=36 m amplitude
We compare our results to a previous study that uses more sophisticated methods to analyze ambient
data acquired from only the subsurface array. Luo et al. (2020) present a detailed 1-D S-wave velocity
profile using a multimodal Monte Carlo inversion method for the first 120 m of the subsurface. We find
that their numerical results suggest a VS30 estimate of approximately 370 m/s for the entire survey area.
This result falls within approximately 5% of our VS30 estimates along the north side of the field, and 9%
toward the south side (Figure 3.5).
3.5 Discussion
Figure 3.2 shows that the time-domain waveforms for all subsections exhibit asymmetric behavior. The
observed effect is an amplitude difference between signals traveling along the positive (causal) and negative
(acausal) time lags. Ideally, these signals would be identical if the sources of ambient energy were
distributed homogeneously throughout the survey area [49]. However, the waveform asymmetry is likely
due to the heavy traffic on the road directly to the west (Figure 3.1). We make this interpretation based
on: (1) the road running perpendicular to the DAS subsections; and (2) the proximity (approximately 5 m)
to the nearest DAS channel. The DAS instrument response is highly sensitive to propagating waves with
particle motion oriented along the fiber axis. Thus, the primary source of energy is likely from the
high-amplitude surface waves generated by nearby traffic noise.
33
|
Colorado School of Mines
|
Figure 3.5 Estimated VS30 results for each linear subsection of the surface array (yellow lines) and
subsurface array (red lines). The colored circles and reported values represent the VS30 estimates for the
entire length of the corresponding subsection of fiber. Lou et. al. (2020) VS30 estimate was calculated
using the entire subsurface array.
The spectral differences presented in our results commonly arise when applying an interferometric
approach to field data [49]. The asymmetry in ambient waveforms is a consequence of wavefield energy
distribution and array geometry. The most effective seismic array geometries for near-surface surveys
require consideration of the spatial distribution of ambient energy sources. Dominant noise sources (e.g.,
roads) should be of particular importance [17]. This is especially true in terms of time and cost if sensors
were trenched prior to data acquisition. The novelty of our surface deployment minimizes the economic
risk involved in both static and time-lapse surveys, allowing for flexible installations and rapid
reconfiguration of seismic arrays if necessary.
Our study indicates that robust VS30 estimates can be made from linear DAS arrays deployed directly
on the ground surface. Further analysis of the ambient waveforms and dispersion curves provides
qualitative information describing the coupling of each subsection. In general, we observe that waveform
34
|
Colorado School of Mines
|
3.6 Conclusions
We demonstrate the effectiveness of DAS to record dynamic strain from surface waves generated by
anthropogenic activity when deployed directly on the ground surface under different coupling conditions.
We deployed a DAS array composed of six parallel linear subsections along the ground surface above
preinstalled collocated trenched cables buried 1 m below. Applying an interferometric approach to 1.0 hour
of recorded data, we show that the resultant waveforms and dispersion curves contain qualitative
information describing the variable coupling of each linear subsection. Coherent high-amplitude waveforms,
moveouts across channels, and continuous dispersion features characterize the coupling quality.
Furthermore, ambient waveform asymmetry suggests that the data are influenced by the spatial
distribution of dominant energy sources, highlighting the importance of flexible seismic array installations.
By adopting a simplified spectral-analysis-of-surface-waves (SASW) method that does not require
inversion, we produce robust VS30 estimates consistent for collocated subsections despite the differences in
coupling condition. Implications of this study suggest that DAS can be rapidly deployed along the ground
surface to acquire high-quality data for a variety of near-surface seismic applications.
3.7 Acknowledgments
The instrumentation used for this study belongs to the DFOS laboratory at Colorado School of Mines.
. We would like to thank Dr. Richard Krahenbuhl and Dr. Whitney Trainor-Guitton for facilitating the
fiber installment of the Mines Underground Laboratory. We would also like to acknowledge Dr. Bin Luo
for his research involving the Mines Underground Laboratory fiber installment which has been cited in this
study. This chapter was converted from a manuscript first-authored by Joseph Mjehovich, and co-authored
by Dr. Ge Jin, Dr. Eileen R. Martin, and Dr. Jeffery Shragge. My contributions in the work include
assisting in the deployment, acquisition, and data processing. I developed the workflows and codes used for
this study including cross coherence and dispersion processing and analysis.
36
|
Colorado School of Mines
|
CHAPTER 4
CONCLUSIONS AND FUTURE WORK
In the preceding chapters we demonstrate the versatility and multi-domain capabilities of DAS through
the practical application of two distinct and novel methodologies. This work spans two very different
industries and audiences: (1) exploration (i.e., oil and gas), and (2) geotechnical. Critical to both is the
availability of cost-effective methods to acquire robust data in logistically challenging settings. We leverage
the exceptional broadband frequency response of DAS, and its high spatial-temporal resolution to address
these challenges. Our developed methodology are computationally efficient, simple to automate and
integrate into existing workflows, and offer an economic alternative to established methods. Although our
results are promising and validate the multi-functional use of DAS in conjunction with our proposed
methodology, we acknowledge the potential for further development. Therefore, we offer several areas of
future work to be considered.
4.1 Conclusions
Chapter 2 demonstrates that DAS can be deployed in an offset well to diagnose multi-stage hydraulic
fracturing treatments in the far-field. We apply a geomechanical inversion algorithm to quantitatively
constrain fracture widths using low-frequency DAS data which can be interpreted to evaluate the degree of
inter-stage fluid communication in an injection well (i.e., poor stage isolation, bridge plug failure, casing
failure). The unique optical-fiber deployment at the DJ-Postle well site provides the opportunity to
compare offset well LF-DAS interpretations with established in-well diagnostic methods (i.e. DAS and
DTS). LF-DAS results indicate incomplete stage isolation in three of four analyzed intervals, which is
validated with in-well DTS analysis. However, in-well DAS interpretations suggest only one of four stages
suffers from poor stage isolation. We conclude that the in-well DAS measurements for the targeted
treatment stages are unreliable due to in-well and near wellbore region erosion effects (i.e., proppant
induced erosion of perforation holes and, near-wellbore fractures).
Perhaps the most significant result is the sensitivity of the inversion algorithm to variations in the
injection well. Notably, we observe abrupt pressure changes in the injection well coincident with fracture
width changes in the offset well at least 250 ft away. Integrating our analysis across all four targeted stages
shows the effects of poor stage isolation on completion efficiency. As inter-stage fluid communication
propagates into adjacent stages, fracture width growth diminishes substantially in the far-field.
37
|
Colorado School of Mines
|
The implications of our results promote the use of LF-DAS to acquire a suite of diagnostic information.
Qualitatively, it can be used to identify inter-well communication in the far-field, constraining fracture
height, length, density, and azimuth. Quantitative constraints can be placed on fracture width
simultaneously providing critical information for in-well diagnostic interpretations to optimize completion
parameters. LF-DAS installations in an offset well can substantially lower the risk associated with costly
and challenging in-well installations for operators and service companies. Finally, our developed method is
computationally efficient and can be automated to handle large scale completion operations.
Chapter 3 demonstrates the viability of DAS to record high-fidelity surface waves deployed directly on
ground surfaces under different conditions. We collected 1.0 hr of ambient data on a surface deployed DAS
array and verified our results with the collocated trenched array 1 m beneath the surface. We applied a
simplified multi-channel analysis of surface waves which includes ambient interferometry resulting in
estimated VS30 across Kafadar Field located on the Colorado School of Mines campus. We find that VS30
estimates result in an approximate 15% spatial variability across the width of the field (45 m).
This field study highlights the exceptional sensitivity of DAS to record surface waves, which is critical
for MASW surveys. Despite relatively light coupling conditions, each section of fiber recorded sufficient
signal response to produce consistent VS30 estimates with only 1.0 hr of data. The cross-coherence and
stacking process associated with ambient interferometry successfully offset the poor signal to noise ratio
synonymous with poor coupling conditions, resulting in coherent waveforms and dispersion spectra.
The developed method has practical application for the greater seismic community (i.e., earthquake
aftershock monitoring, regional scale imaging). However, geotechnicians would likely benefit the most, as
our method could possibly be implemented into existing MASW workflows used for site classification
surveys. An additional benefit is that DAS surface deployments are low impact, and do not require
negotiation of challenging installations (i.e., burying the cable, downhole measurements). Finally, the use
of DAS among the geotechnical community would significantly impact exposure and commercialization,
further advancing a rapidly developing technology.
4.2 Future Work
4.2.1 Low-frequency DAS for Diagnosing Multi-stage Hydraulic Fracture Treatments
Quantitative analysis using in-well measurements to estimate fluid flow volumes would be the next
logical step in the development of this method. DAS and DTS fluid flow modeling could be used to verify
our results beyond qualitative analysis. Placing constraints on the fracture geometry recorded at an offset
well would enable far-field fluid volume estimates to be compared with in-well modeling, possibly
establishing a relationship between the two. Ideally, fractures measured at multiple offset wells could be
38
|
Colorado School of Mines
|
used to reduce the amount of unknown parameters and uncertainty associated with the fluid volume
measurements.
Further automation of the workflow would permit large-scale diagnostics for hydraulic fracturing
treatments. Qualitative analysis (i.e., fracture length, height, azimuth) coupled with quantitative (i.e.,
fracture width) constraints could provide unprecedented information critical to optimizing completion
design and well spacing. The availability of this data for an entire well completion would provide the
information needed to make geomechanic interpretations complementary to current methods used for
reservoir characterization.
4.2.2 DAS Surface Deployment
During our analysis we found that the the Grass Pressed and Cement South subsections returned peak
spectral values resulting in VS30 estimates far outside the range of the other subsections. After further
analysis, we found that the second highest peak spectral value resulted in VS30 values consistent with the
other subsections. We postulate that this is likely due to the relatively sparse frequency resolution from our
dispersion processing. Although we applied interpolation to enhance the resolution, perhaps a finer or more
advanced interpolation should be applied to address this issue. The quality of the Cement South subsection
was relatively poor, thus interpreting the second highest spectral peak can be argued. However, the
dispersion curve for the Grass Pressed was much higher quality, leading to uncertainty in our
interpretations. This is an important issue to address because if this method was applied in the field, the
deployment configuration may not have multiple subsections to aid with interpretations.
There is a notable difference between the dispersion curves observed for Cement South and Cement
North, with the former lacking any coherent dispersion features above 9 Hz. The deployment of both
subsections was identical outside of the location in the study area. It is possible that the observed effect is
caused by noise from the nearby CoorsTek building. Cement South is the subsection closest to the
building, and is also the tail end of the array which was deployed inside CoorsTek where it was connected
to the subsurface array.
Calculating V models using more sophisticated inversion methods on each subsection and comparing
s
with V models from the subsurface array would be the next step to verify our results. V model
s s
comparison using traditional methods (i.e., geophone deployment) would likely resonate with the general
seismic community on a greater scale.
39
|
Colorado School of Mines
|
APPENDIX A
CHAPTER 1 SUPPLEMENTAL INFORMATION
A.1 Perforation and Plug Depth Location
Determination of the perforation and bridge plug depth locations is critical for diagnosing stage
isolation and the degree of inter-stage fluid communication. Depth information provided by Great Western
Petroleum is used as a first-degree initial estimate. We plot the in-well DAS data for the four targeted
stages (Figure A.1 top), including the temporal acoustic energy summation (Figure A.1 bottom), using the
500-5000 Hz frequency band which was preprocessed by the service provider. The y-axis indicates time
(ascending top to bottom) and the x-axis indicates the measured depth along the wellbore (heel to toe).
The dashed black lines represent the estimated perforation locations. Assuming a correlation between
acoustic intensity and fluid flow, events recorded by in-well DAS can be used to align the measured depths
between perforations and DAS channels.
Qualitative observations provide an approximate location; however, it is difficult to distinguish
individual perforations. We applied several bandpass filters to better identify assumed perforation noise,
but the differences were negligible for determining individual perforations. The temporal energy
summation (Figure A.1 bottom) provides the energy profile along the stage intervals, where the peak
amplitudes are assumed to align with perforation locations. We used the high amplitude peaks recorded at
the toe-side clusters as our proxy and apply a 10 ft bulk shift to the perforation depth locations. We make
this interpretation based on the following:
• Not all perforations align with a peak amplitude in the summation plots (Figure A.1 bottom).
Specifically, for 12 perforations, we would ideally observe twelve well defined amplitude peaks
separated by troughs. This absence is possibly due to uneven fluid distribution across the
perforations, and/or near-well bore region effects (i.e., fractures along the wellbore).
• The perforation(s) at the toe-side of the targeted stage interval record the highest acoustic intensity.
This is likely due to positional bias, with toe-side clusters receiving a greater amount of proppant.
Deneshy [11] observed separation between proppant and fluid, with an excess of proppant traveling to
the toe-side clusters due to inertia.
We also compute the RMS amplitude (see section 2.4) to support our interpretations. Figure A.2 shows
the RMS amplitudes, plotted against depth, for each treatment stage. The black dots indicate the
perforation locations and are shifted along the y-axis to visually differentiate between stages. Applying the
44
|
Colorado School of Mines
|
Figure A.2 RMS amplitude calculated from in-well DAS (500 - 5000 Hz) for stages 17-20. The black dots
indicate the perforation measured depth locations. Offset in the y-direction is to differentiate stages.
10 ft bulk shift in depth aligns the perforations with the high amplitude toe-side signals and confines them
within the stage intervals.
A.2 In-well DAS Acoustic Intensity Attribute
The static in-well DAS attribute (see section 2.4) serves as a semi-quantitative method to characterize
the total acoustic intensity recorded within the targeted, and adjacent treatment stages. Figure A.3 shows
the acoustic intensity recorded during treatment stage 20 as an example. The dashed black lines indicate
the bridge plug depths, and the red dashed lines indicate the depth interval used to calculate the
“noise-floor”. The process is as follows:
• Window data spatially according to plug set depth (i.e. between the black dashed lines).
• Window data temporally according to the start and end time of the treatment.
• Compute the absolute sum (i.e., sum across the spatial and temporal axis of the data).
• Repeat the process to compute the noise-floor for each treatment stage using an upstream depth
interval with relatively little noise.
46
|
Colorado School of Mines
|
A.3 LF-DAS pre-inversion Processing
It is necessary to preprocess the LF-DAS data prior to inversion to avoid unreliable fracture width
results. Note that preprocessing steps may vary based on the quality of data (i.e., background noise,
temporal resolution, interrogator used for acquisition). The challenge is to filter the data while minimally
impacting the targeted extension and compression signal(s) induced by fracture hits. The first step in our
process is defining a data window upstream from the targeted treatment interval. The upstream interval is
preferred because we assume the formation has not been affected from any downstream treatment stages in
the injection well. The median is calculated along the temporal axis for each DAS channel in the selected
window and subtracted from the entire LF-DAS dataset to remove the low-frequency background drift
noise that is most likely associate with interrogator noise [4]. We then apply a median filter (3x5 kernel) on
the entire dataset to remove the “spikey noise” (Figure A.4) which is commonly observed in LF-DAS data
(e.g., [4][14]). We postulate this noise is due to mechanical decoupling as the formation extends and
compresses spatially along the fiber.
To aid in the fracture picking process, we extend the LF-DAS data by a sufficient amount of time prior
to injection and after injection ends. In Figure A.4 we observe previous stage fractures closing prior to
injection. After injection ends, we observe a polarization in the strain signal which is a key indicator of the
fracture(s) measured depth location.
A.4 Field Strain vs Model Prediction
Figure A.5 show the LF-DAS waterfall plots including the field strain change used for inversion
(Figure A.5a and Figure A.5d) and the predicted model strain using the inverted widths (Figure A.5b and
Figure A.5e) for stages 17 and 18. The field strain and model predicted strain match well suggesting
confidence in our inversion results. For a more straightforward comparison, Figure A.5c and Figure A.5f
show the field strain and model predicted strain for a specific timestep near the end of the treatment for
stage 17 and 18, respectively. Figure A.6 shows the same results for stages 19 and 20. In general, all stages
show good agreement between the field and predicted strain.
The extension strain data at the fracture hit location is removed prior to inversion. LF-DAS data may
not accurately represent fracture-induced strain at the fracture hit location due to mechanical decoupling
and thermal effects [12]. When the fracture approaches the fiber, the separation between the
cement/formation and the fiber may cause measurement bias, impacting the accuracy of the inversion
results. Thermal effects may also result from direct hydraulic fracturing fluid contact with the fiber,
introducing additional bias.
48
|
Colorado School of Mines
|
APPENDIX B
CHAPTER 2 SUPPLEMENTAL INFORMATION
B.1 Spatial Coupling Variation Example
Ambient waveform static results (Figure 3.2) provide qualitative insights regarding the coupling
condition at one virtual source (DAS channel) for each subsection. However, the interformetric process (see
section 3.3) results in a 3D volume of waveforms, where each DAS channel acts as a virtual source and
receiver. Scanning through each DAS channel provides a quick, qualitative analysis of how the coupling
condition varies across the entire field for each subsection.
The Snow Pressed and Snow subsections provide an intuitive example of how coupling conditions can
effect the quality of the recorded signal. Figure B.1 is a map view of Kafadar Field, indicating the
approximate location and direction of DAS channels (in ascending order).
Figure B.1 Map view of Kafadar Field. The purple and red lines indicate the location and ascending
channel direction of the Snow Pressed and Snow subsections, respectfully.
Figure B.2 shows the ambient waveforms recorded along the Snow Pressed subsections at four different
DAS channels. In general, we observe a strong signal response indicated by high amplitude waveforms and
moveouts for each of the DAS channels. Coherent signals are strongest closer to the road, and begin to
diminish as the virtual source moves toward the Green Center.
In contrast, Figure B.3 shows that the Snow subsection records very little coherent waveforms for most
of the virtual sources. This indicates insufficient coupling with the Snow for DAS to record any notable
surface wave signals. However, as the virtual source becomes closer to the road, we observe some coherency
52
|
Colorado School of Mines
|
Figure B.2 Ambient waveforms recorded along Snow Pressed subsection showing how coupling varies
spatially across Kafadar Field. From west to east: (a) channel 273, (b) channel 287, (c) channel 296, and
(d) channel 306.
and moveout in the waveforms. Assuming continuous coupling along the Snow subsection, this
demonstrates the sensitivity of DAS to record surface waves despite relatively poor coupling.
Several important conclusions can be made from this example alone:
• Light coupling (i.e. pressing fiber into the ground) can provide enough fiber-ground contact to
leverage the high sensitivity of DAS to record surface waves.
• Surface deployments with DAS require careful consideration of array orientation and proximity to
ambient sources in the survey area. The process presented in this study may overcome the challenges
of unfavorable coupling by simply deploying fiber closer to dominant energy sources.
• Despite the challenge of incoherent signal for most of the Snow subsection, the process presented in
this study produced consistent VS30 results compared with the other subsections. This suggests that
as little as a few tens of meters of sufficient coupling can be used for near-surface surveys (i.e. VS30
assessments).
• Our methodology can be modified and automated to make quick in-the-field assessments. This allows
technicians to quickly scan the data (i.e. scan through virtual sources) and determine if array
53
|
Colorado School of Mines
|
ABSTRACT
Valves are widely used in transportation systems for various reasons. In oil and gas pipelines, they
regulate the flow rates or pressures to meet facility requirements or ensure safe transportation. However,
their effect on multiphase flow behavior is still not well understood. A major purpose of this study is to
systematically investigate the choking effect on downstream multiphase flow behaviors for three-phase slug
flow. To achieve this goal, a flow loop consisting of a 43-ft long horizontal pipe with a 2.067-in. inner
diameter was constructed, with a 2-in. ball valve installed at the inlet of the test section. Flow patterns,
phase distributions, and pressure drop were measured at the test section 123 pipe diameters downstream of
the valve. The instrumentation includes a high-speed camera, an Electrical Capacitance Volume
Tomography system for phase distribution monitoring, and differential pressure transducers for pressure
drop measurements in the test section and across the valve. A total of 67 tests for oil-water two-phase flow
and 106 tests for gas-oil-water three-phase slug flow were conducted to systematically investigate the effects
of the inlet choke opening, water cut, gas, and liquid superficial velocities on the downstream fluid flow
behaviors.
Based on the experimental observations, new models were proposed for oil-water and gas-oil-water flows
in horizontal pipes, respectively. The model for oil-water flow focuses on semi-dispersed flow that
demonstrates a dispersion layer on top with a free water layer at the bottom. The model for gas-oil-water
three-phase flow focuses on slug flow, in which the slug body and film region demonstrates some
stratifications in the liquid phase. Parametric studies show that the new three-phase slug flow model
captures well the effects of choking opening, water cut, liquid and gas flow rates on the pressure gradient.
Both models outperform other existing models.
In summary, this study provides valuable insights into how an inlet choke affects the downstream fluid
flow behavior for oil-water two-phase and gas-oil-water three-phase flows in horizontal pipes at different
flowing conditions. It also offers two new hydraulic models for predicting phase holdup and pressure
gradient with a focus on flow patterns that lack deep understanding.
iii
|
Colorado School of Mines
|
CHAPTER 1
INTRODUCTION
Valves are typical restrictions that are widely used in the oil and gas industry, especially for
transportation pipelines. Their ubiquitous applications are primarily due to the necessity to control the
flow rates or pressures to meet facility requirements or ensure safe transportation. Their presence can
significantly affect fluid flow behaviors, particularly in the case of multi-phase flow. Gas-oil-water
three-phase flow frequently occurs with the maturity of an oil field since water production is almost
inevitable. This is even more common in offshore scenarios because of the limited space to accommodate
all the necessary facilities for a complete separation. Therefore, understanding the effects of choke on the
downstream fluid flow behaviors is of great significance.
Though liquid-liquid and gas-liquid two-phase flows have been extensively studied, gas-liquid-liquid
three-phase flow is still not well understood. In particular, the influence of choke on downstream fluid flow
behaviors is rarely investigated. One of the major objectives of this study is to understand the effects of
choke openings on the downstream flow behaviors in three-phase flow. Our focused flow pattern is
three-phase slug flow due to its complexity and lack of studies. To achieve this goal, a flow loop with a 43-ft
long horizontal test section was built in the laboratory to carry out experimental studies. A 2-in. nominal
ball valve was installed at the inlet to examine the effects of choke openings on downstream flow behaviors.
Gas-oil-water three-phase flow experimental work was conducted at various inlet choke openings, water
cuts, and flow rates. Additionally, respective oil-water two-phase flow experiments were conducted to
better understand their relationships with the gas-oil-water three-phase flow. Moreover, existing hydraulic
models were evaluated, and new models were proposed based on the findings of the experimental work.
The structure of this thesis is: Chapter 1 provides an overview of the background, goals, and thesis
structure. Chapter 2 presents literature review related to the topic of this thesis. The current progress of
three-phase flow studies in horizontal pipes is also reviewed, including experimental studies and modeling
studies. Chapter 3 outlines the objectives and tasks. Chapter 4 focuses on introducing experimental
facilities and analyzing our experimental results of oil-water two-phase and gas-oil-water three-phase flows.
Various inlet choke openings are a major consideration in those experimental studies. Chapter 5 discusses
new model development and model evaluation. Lastly, in Chapter 6, the key findings are summarized and
recommendations for future studies are briefly presented.
1
|
Colorado School of Mines
|
CHAPTER 2
LITERATURE REVIEW
This chapter provides a literature review in terms of experimental studies on three-phase flow,
experimental studies on fluid flow in pipes with restrictions, gas-liquid two-phase slug flow modeling, and
gas-oil-water three-phase flow hydraulic modeling in horizontal pipelines.
2.1 Previous Experimental Studies on Three-phase Flow in Horizontal Pipes
In this section, previous experimental studies centering on flow patterns, liquid holdup, and pressure
drop of three-phase flow in horizontal pipelines are reviewed.
2.1.1 Flow Pattern
This subsection reviews the reported flow patterns in gas-liquid-liquid three-phase flow, with a focus on
how flow pattern is defined in this context. Critical studies are reviewed in the following.
Acikgo¨z et al. (1992) conducted experiments in 19 mm inner diameter plexiglass tubes and
systematically investigated the flow regimes of three-phase flow in horizontal pipelines by varying both the
superficial water and air velocities at a constant superficial oil velocity. The used superficial water velocity
ranges from 0.4 to 60 cm/s, and the range of superficial gas velocity is 14.22-5000 cm/s. They
discriminated plug flow from slug flow by the driving phase (it is plug flow if the liquid is the driving
phase, and it is slug flow if the gas is the driving phase) in classifying the flow patterns. The nomination
rule for the flow pattern begins with the base phase (the phase with the largest volume fraction) followed
by the liquid-liquid flow pattern and then the gas-liquid mixture flow pattern, though the rule was not
explicitly introduced. It’s worth mentioning that a free water layer is present at the bottom of the pipe for
flow-pattern numbers 3, 4, 5, and 6 (Figure 2.1), which still fall into the oil-based flow patterns. This
means that the oil-based and water-based flow pattern is not differentiated based on whether oil or water is
the continuous phase. In summary, they classified six oil-based flow patterns (1-dispersed plug flow,
2-dispersed slug flow, 3-dispersed stratified/wavy flow, 4-separated stratified/wavy flow, 5-separated wavy
stratifying-annular flow, and 6-separated/dispersed stratifying-annular flow) and four water-based flow
patterns (7-dispersed slug flow, 8-dispersed stratified/wavy flow, 9-separated/dispersed incipient
stratifying-annular flow, and 10-dispersed stratifying-annular flow), as shown by Figure 2.1. Note, j , j ,
a o
and j denote superficial velocities for air, oil, and water, respectively.
w
2
|
Colorado School of Mines
|
Figure 2.1 Flow pattern map for three-phase flow at a superficial oil velocity of 9.0 cm/s in a horizontal
pipe (Acikg¨oz et al., 1992).
Lee et al. (1993) conducted experimental work to observe flow patterns for water-oil-carbon dioxide
three-phase flow in a 10 cm inner diameter horizontal plexiglass pipeline. The liquid and gas velocity
ranges are 0.05-2 m/s and 0.5-15 m/s, respectively. For gas-oil-water three-phase flow, they defined in a
similar way as those observed in gas-liquid two-phase flow and identified seven flow patterns, including
smooth stratified, wavy stratified, rolling wave, plug flow, slug flow, pseudo slug, and annular flow. They
also mentioned that those flow patterns are combinations of flow patterns in gas-liquid and oil-water
two-phase flow.
With regard to slug flow, they pointed out that the oil flows above water at low gas and low liquid flow
rates. However, oil-in-water mixture forms when either the gas or liquid flow rate is increased. The mixture
is not homogeneous with more oil close to the top of the pipe and more water settling down to the pipe
bottom. But, the liquid mixture will be well-mixed at high gas velocity.
Pan (1996) measured three-phase flow patterns and found a major drawback of the name rule used by
Acikg¨oz et al. (1992) in cases where oil and water flowed separately, and the flow pattern could not be
named as either oil-based or water-based. As such, he introduced a new three-part naming convention to
classify three-phase flow patterns. The first component of the name depicts the relationship between oil
and water, the second component identifies the continuous phase in the liquid mixture, and the third
component represents the flow pattern of air-liquid. It is noteworthy that the continuous phase is not
included in the name when the oil-water demonstrates a separated flow. He conducted experimental work
in a horizontal or 1-degree upward 3-in. steel pipeline. The testing fluids are Shell Tellus 22 oil, tap water,
3
|
Colorado School of Mines
|
and air. The superficial velocities for air, oil, and water are 0-25.87, 0-0.565, and 0-0.615 m/s, respectively.
He observed eight flow patterns, including separated stratified (SSt), dispersed oil continuous stratified
(DOSt), dispersed water continuous stratified (DWSt), separated slug flow (SSl), dispersed water
continuous slug flow (DWSl), dispersed oil continuous slug flow (DOSl), dispersed oil continuous annular
(DOA), and dispersed water continuous annular (DWA).
Spedding et al. (2005) conducted three-phase flow experiments in both 0.0259 m and 0.0501 m inner
diameter horizontal pipelines. The testing fluids are Oil, water, and air. They first divided the flow
patterns into either oil-dominated (Figure 2.2) or water-dominated (Figure 2.3). Then, they used stratified,
intermittent, and annular to further classify the three-phase flow patterns. Though they did not explicitly
introduce the flow pattern designation rule, it seems to start with gas-liquid flow patterns followed by
liquid-liquid flow patterns after the discrimination between oil-dominated and water-dominated ones. They
sketched 22 flow patterns, as shown in Figure 2.2, Figure 2.3, and listed in Figure 2.4, but only observed
six of them in their experiments based on their reported flow pattern map.
Figure 2.2 Oil-dominated flow patterns in three-phase horizontal pipe flow adapted from Spedding et al.
(2005).
Wegmann et al. (2007) captured flow patterns of three-phase pipe flow for two distinctive pipe inner
diameters (5.6 and 7mm) by high-speed photography. The testing fluids are paraffin oil, deionized water,
and air. They blended a fluorescence dye into the water phase and employed a laser-induced fluorescence to
light the pipe for photography. They explicitly defined their flow patterns based on a two-part rule, with
the first part describing the liquid-liquid flow and the second part depicting the gas-liquid flow. They
observed six flow patterns, namely, stratified-intermittent, annular-intermittent, intermittent-dispersed,
4
|
Colorado School of Mines
|
patterns into stratified, dual continuous, water continuous, and oil continuous, and gas-liquid flow patterns
into stratified, intermittent, annular, and dispersed bubble flow. Finally, they came up with twelve flow
patterns for three-phase flow. Additionally, they experimentally examined three-phase flow patterns in a
horizontal pipeline of 0.0508m inner diameter with mineral oil, tap water, air, and a high-speed camera.
The superficial velocity ranges for air, oil, and water are 0.1-7.0, 0.02-1.5, and 0.01-1.0m/s, respectively.
They observed eight flow patterns for three-phase flow in the horizontal pipeline with the absence of
annular and dispersed bubble flow for gas-liquid flow. Six observed flow patterns are shown in Figure 2.5,
namely, stratified-stratified (ST-ST), stratified-oil continuous (ST-OC), stratified-dual continuous
(ST-DC), intermittent-stratified (IN-ST), intermittent-oil continuous (IN-OC), and intermittent-dual
continuous (IN-DC). The other two flow patterns are stratified-water continuous(ST-WC) and
intermittent-water continuous (IN-WC) which were observed at a water cut equal to 50% or higher. The
corresponding flow regime map is not listed here for simplicity.
Figure 2.5 Flow pattern map at a water cut of 40% for three-phase flow in a 0.0508 m inner diameter
horizontal pipe (Keskin et al., 2007).
Bannwart et al. (2009) investigated gas-oil-water flow patterns in a 2.84 cm inner diameter steel
horizontal pipeline with viscous crude oil, and identified nine flow patterns, as shown in Figure 2.6. They
defined flow patterns starting with the gas-water flow patterns followed by the oil-water flow patterns.
Since a major motivation of their study is to investigate heavy oil transportation under core-annular flow
conditions, almost all observed flow patterns are water continuous.
Wang et al. (2013) investigated the three-phase flow using natural gas and viscous oil in a horizontal
pipeline. They adopted a two-part combined nomination rule describing the gas-liquid followed by
6
|
Colorado School of Mines
|
liquid-liquid interactions. In particular, they further sub-classified the intermittent flow pattern into INT
(O/W-S & SOW-F) as shown by Figure 2.7, INT (O/W-S & O/W-F), and INT (W/O-S & W/O-F) based
on the phase distribution both in the slug and film regions. The first one denotes an oil-in-water dispersion
in the slug body and oil-water stratified flow in the film region. The second one represents oil-in-water
dispersion in both regions. The last one indicates a water-in-oil dispersion in both regions. All those three
flow patterns have a thick oil film in the film region.
Figure 2.6 Viscous crude oil-water-gas flow patterns (Bannwart et al., 2009).
Kee (2014) used a high-speed camera to observe the flow pattern of three-phase flow in a 4-in. inner
diameter horizontal PVC pipeline. The testing fluids are paraffinic oil, 1 wt.% NaCl water, and carbon
dioxide as the gas phase. The superficial velocity ranges for the liquid mixture and gas are 0.2-1.5 and 1-45
m/s, respectively. The water cut is 1-20%. The author identified five oil-continuous flow patterns in a
horizontal pipe, namely stratified flow, elongated bubble (i.e., plug flow), slug flow with a stratified liquid
layer in the film region, wavy annular flow, and annular-mist flow. These flow patterns are defined similarly
7
|
Colorado School of Mines
|
to gas-liquid flow patterns. However, it should be noted that both stratified and elongated-bubble flows are
not truly oil-continuous as the author claimed, given the presence of a free water layer at the pipe bottom
as shown in the author’s flow pattern sketch.
Figure 2.7 INT (O/W-S & SOW-F) flow pattern in three-phase horizontal pipeline flow. (a) slug body
region, (b) liquid film region (Wang et al., 2013).
Husein et al. (2021) systematically examined the waxy crude oil-water-air three-phase flow patterns in
a 0.041m inner diameter horizontal pipeline. They defined the three-phase flow pattern with a combination
of gas-oil and oil-water flow patterns and identified eight flow patterns, shown in Figure 2.8.
Figure 2.8 Waxy crude oil-water-gas flow pattern map in horizontal pipelines (Husein et al., 2021).
In summary, the flow pattern of gas-liquid-liquid three-phase flow is much more complicated than
gas-liquid or liquid-liquid two-phase flow. Generally, gas-liquid-liquid flow shows the same flow patterns as
in gas-liquid flow if the flow pattern of the liquid-liquid phase in three-phase flow is not considered. Indeed,
some researchers did define the flow pattern of gas-liquid-liquid three-phase flow in almost the same way as
that for gas-liquid two-phase flow (Lee et al., 1993; Kee, 2014). Nevertheless, the available literature
suggests that most studies use a two-part or three-part nomination rule. The former is to combine the flow
8
|
Colorado School of Mines
|
patterns of liquid-liquid flow with those of gas-liquid flow to characterize three-phase flow patterns (Pan,
1996; Wegmann et al., 2007; Keskin et al., 2007; Bannwart et al., 2009; Husein et al., 2021). The latter
follows the same rule as that for the two-part one but with an additional part to define the continuous
phase for the liquid-liquid mixture (Acikgo¨z et al.,1992; Spedding et al., 2005). It is worth pointing out
that most of the aforementioned literature uses low-viscosity oils while few studies (Bannwart et al., 2009;
Wang et al., 2013) deal with viscous oil flow. Wang et al. (2013) additionally depicted the liquid-liquid
dispersion states in both the slug region and the film region for three-phase slug flow.
2.1.2 Liquid Holdup and Pressure Gradient
Liquid holdup and pressure drop are two crucial parameters in multiphase flow, and they are closely
related to many factors, such as pipe geometry, system pressure, superficial velocities, water cut, and fluid
properties. This subsection reviews the effects of these factors on both liquid holdup and pressure gradient
in three-phase flow.
2.1.2.1 Liquid Holdup
Liquid holdup in horizontal pipe flow is primarily affected by superficial velocity, liquid viscosity, and
pipe friction (Husein et al., 2021). Other factors, including system pressure and water cut, can also have an
impact on the liquid holdup. This part reviews the effects of superficial velocity, system pressure, and
water cut on liquid holdup.
Effect of superficial velocity
In general, liquid holdup decreases with increased superficial gas velocity but increases with a growing
superficial liquid velocity (Husein et al., 2021), shown by Figure 2.9. The highest liquid holdup was
observed at the highest superficial liquid velocity and the lowest superficial gas velocity when the flow
pattern was dispersed bubble-oil continuous flow. The lowest liquid holdup was observed at slug-oil
continuous flow pattern, where superficial gas velocity was high and superficial liquid velocity was low.
Effect of system pressure
System pressure is believed to greatly influence the gas phase density and thus can affect liquid holdup
(Pan, 1995). Pan (1995) observed a smaller liquid holdup at a higher system pressure. He measured liquid
holdups for various water cuts at a superficial gas velocity of 2 m/s and a superficial oil velocity of 0.1 m/s.
the liquid holdup is higher for the case at a 0 bar pressure than that of a pressure of 10 bar. It is worth
mentioning that the flow pattern also changed with increasing system pressure. The flow pattern changed
from DOSl to DWSl with the water cut increase at a 0 bar system pressure, while it was SSt for all the
tested water cuts at a 10 bar system pressure.
9
|
Colorado School of Mines
|
Figure 2.9 Liquid holdup of waxy crude oil-water-gas flow in horizontal pipelines (Husein et al., 2021).
Effect of water cut
Pan (1995) investigated the effect of water cut on liquid holdup in gas-oil-water three-phase flow in
both horizontal and one-degree upward inclined pipelines. We only reviewed the results for horizontal slug
flow at atmospheric pressure conditions, which are more relevant to the current study. He measured the
liquid holdup as a function of water cut at different superficial gas velocities and a fixed superficial oil
velocity using a dual-energy gamma densitometer system. He concluded that the water holdup first
increased, then dropped at the inversion point, and increased again with a further increase of the water
cut. By contrast, the oil holdup had a continuously decreasing trend with the water cut. The decrease was
gradual at lower water cuts before the inversion point and became much faster close to the inversion
region. As such, the total liquid holdup first increased with the water cut until reaching a peak, then
dropped to a minimum, and slightly increased with a further water cut increase. He believed the first
increase was due to the growing viscosity with the water cut in the oil continuous region before the
inversion point, and the second increase was because of the elevated water flow rate. In addition, the total
liquid holdup decreased with an increased gas flow rate, and the peak of the total liquid holdup slightly
moved to a higher water cut at a higher gas flow rate.
Wang et al. (2013) examined the effect of water cut on viscous oil-water-natural gas flow in a 5.25 cm
ID horizontal pipeline. The used mineral oil has a viscosity of 0.15 Pa·s at 37.8 °C, and a surface tension
of 35.75 mN/m at 19.8 °C. The dominant flow pattern is slug flow with oil dispersed in water both in the
slug and film regions. The oil holdup showed a decreasing trend with increasing the water cut, while the
water holdup increased with the water cut (Figure 2.10).
10
|
Colorado School of Mines
|
Figure 2.10 Measured oil holdup (left) and water holdup (right) as a function of the input water cut in
gas-oil-water three-phase slug flow (Wang et al., 2013).
2.1.2.2 Pressure Gradient
The pressure gradient of horizontal pipe flow is mainly influenced by the flow pattern, flow rates, and
fluid properties (Husein et al., 2021). Regarding the flow pattern, it is generally accepted that pressure
gradient increases with flow pattern transition from stratified to intermittent, to dispersed bubble, and to
annular flow. Note, the flow pattern is also influenced by the flow rates and fluid properties. Pipe diameter
can also impact the pressure gradient when the other flow conditions are the same. The following reviews
the effects of pipe diameter, superficial velocities, and input water cut on the pressure gradient.
Effect of the pipe diameter
Spedding et al. (2008) measured the pressure gradient of gas-oil-water three-phase flow in 0.0259 and
0.0501 m inner diameter horizontal pipelines. They observed an increase in pressure gradient with a
decreased pipe diameter. Despite using an oil with a smaller viscosity (12.2 mPa·s) for the fluid flow study
in the 0.0259 m inner diameter pipe, the measured pressure drop was much larger in this pipe than that
measured in a 0.0501 m inner diameter pipe flowing with a higher viscosity (39.5 mPa·s) oil. It should be
noted that such a result is observed in stratified flow with ripple and rolling waves at a superficial gas
velocity of 10 m/s.
Effect of superficial velocities
Poesio et al. (2009b) measured the pressure gradients of oil-water-gas three-phase flow in a 21 mm ID
horizontal pipeline with very viscous oil. The oil has a density of 886 kg/m3 and a viscosity of 1.2 Pa·s at
20◦C. The superficial gas velocity was fixed at 0.29 m/s. They concluded that the pressure gradient
increased with increasing superficial water or oil velocity (Figure 2.11). It is worth mentioning that the
dominant flow pattern is core annular flow with elongated air bubbles, which aims to better understand its
11
|
Colorado School of Mines
|
Dehkordi et al. (2019a) conducted three-phase slug flow experiments with viscous oil in a 4 cm inside
diameter horizontal pipeline. They concluded that increasing the superficial liquid velocity for a given
superficial gas velocity would increase the pressure gradient. Such an increase was more dramatic for a
lower superficial gas velocity. They believed that this was caused by a transition from the slug to plug flow
(Figure 2.12).
Husein et al. (2021) conducted experiments to understand the effect of water salinity on waxy crude
oil-water-gas three-phase flow in a 0.041 m inner diameter horizontal pipeline. They concluded that the
pressure gradient increases as the superficial gas velocity increases while the superficial liquid velocity
remains constant. Additionally, the pressure gradient is higher for a larger superficial liquid velocity when
the superficial gas velocity is held constant. The effect of superficial gas velocity is more apparent at a
higher superficial liquid velocity (Figure 2.13).
Figure 2.13 Pressure gradients as a function of superficial velocities for three-phase flow (Husein et al.,
2021).
Effect of input water cut
Stapelberg and Mewes (1994) measured the pressure gradient of three-phase slug flow in horizontal
pipelines. They observed that the pressure gradient had a two-region characteristic. In the first region for a
lower superficial gas velocity, the pressure gradient of three-phase flow lay in between those of gas-oil and
gas-water flows. It increased with oil fraction for a fixed superficial gas velocity. In contrast, the pressure
gradient for three-phase flow fell below the gas-water two-phase pressure gradient at higher superficial gas
velocities. They also concluded that the pressure gradients in this region were independent of the oil
fraction for a given liquid flow rate. Additionally, it is apparent that the pressure gradient of three-phase
13
|
Colorado School of Mines
|
Wang et al. (2013) observed an inversion point at 20% water cut for the respective oil-water mixture of
the viscous oil-water-natural gas three-phase flow in a horizontal pipeline. They concluded the three-phase
pressure gradient showed an increasing trend with the water cut increment in the oil-continuous region and
reached a peak close to the inversion point. After that, the pressure gradient decreased again to a
minimum and increased again with a further water cut increase (Figure 2.15).
In summary, various effects on both liquid holdup and pressure gradient have been investigated for
three-phase slug flow in horizontal pipelines. The effects of pipe diameter, system pressure, superficial
velocity, and water cut have been examined. However, it is not well studied considering that there are only
three experimental studies found in the literature (Odozi, 2000; Wang et al., 2013; Dehkordi et al., 2019a)
about three-phase slug flow with two of them using heavy oil (Table 2.1). In particular, none of these
studies have examined the inlet choking effect for three-phase slug flow.
In the following section, we will review studies that specifically address fluid flow through restrictions.
2.2 Previous Studies on Horizontal Pipe Flow with Restrictions
Restrictions commonly exist in the oil and gas transportation system, and various types of valves are
employed to achieve different goals. In terms of chokes, the needle and seat choke, multiple orifice choke,
and fixed-bean choke are typical ones that have a common feature of significant pressure drop across a
short length (Kwakernaak et al., 2007). Chokes are typically mimicked by a circular orifice in a circular
pipe (Khor et al., 1997; Van der Zande et al., 1999; Malot et al., 2003). Various studies have been
conducted investigating their effects on fluid flow behavior. For example, Fossen and Schu¨mann (2017)
conducted experimental work to investigate the influence of a butterfly valve on downstream droplet size
distribution. Some studies investigated other types of valves, such as globe valve (Paolinelli et al., 2018),
gate valve (Silva et al., 2019), and ball valve (Shmueli et al., 2018 and Skjefstad et al., 2020). All those
studies fall into two categories: one aims to study the pressure drop across the restriction, and the other
mainly focuses on its effects on downstream flow behaviors including flow pattern, pressure drop, and
droplet size, as summarized by Table 2.2. Those are reviewed in the following sections.
15
|
Colorado School of Mines
|
2.2.1 Studies of Fluid Flow through Restrictions
The flow through an orifice restriction can be depicted in Figure 2.16 (Vander Zande et al., 1998 and
Charlafti et al., 2021). Due to the cross-section area reduction by the restriction, the fluid accelerates to a
maximum velocity after passing the restriction where the pressure comes to the minimum downstream.
This corresponds to a maximum pressure drop (∆P ) with reference to the upstream pressure. This is
max
also called the vena contracta with a minimum flowing area that is smaller than the orifice area Perry,
2019. It is believed most of the energy is dissipated downstream around the middle zone close to the jet,
based on which the average energy dissipation rate (Equation 2.1) is derived by Van der Zande and Van
der Broek (1998). Recirculation zones also form around the wall due to the low pressure in the middle
zone. At a further position downstream of the restriction, the flow decelerates which comes with an
ensuing pressure recovery, based on which the permanent pressure drop (∆P ) can be defined.
perm
Typically, pressure drop across a restriction corresponds to ∆P at default. The pressure drop
perm
downstream of an orifice restriction was measured by Galinat et al. (2005) with single-phase water flow,
which elaborated the pressure drop characteristics downstream of the restriction very well, as shown in
Figure 2.17. L is typically treated as 2.5 pipe diameter or larger. Morrison et al. (1993) conducted
dis
experimental work to research the turbulence flow downstream of an orifice. They observed that the flow
reattachment occurred 5.3 pipe radii downstream of the orifice, and the pressure fully recovered 17 pipe
radii downstream of the orifice.
∆P V
(cid:15)¯= perm o (2.1)
ρ L
m dis
Perry (1984) proposed a correlation to calculate the maximum pressure drop downstream of an orifice,
given in Equation 2.2. Also, the author proposed to use 1−γ2 to account for the pressure recovery
downstream of the vena contracta which gave an equation for the permanent pressure drop in Equation
2.3. The restriction ratio (γ) is defined as the ratio of the inner diameter corresponding to the opening area
of a restriction to that of the pipe. Galinat et al. (2005) fitted Equation 2.2 and found that the discharge
coefficient varied between 0.8 and 0.9 for a restriction ratio of 0.5-1 and was 0.7 for a restriction ratio of
0.33 independent of the Reynolds number.
1 ρV2(1−γ4)
∆P = p (2.2)
max C2 2 γ4
D
∆P =∆P (1−γ2) (2.3)
perm max
18
|
Colorado School of Mines
|
velocity of sound in that fluid at the in-situ condition. An important feature of critical flow is that the flow
rate across the choke only depends on the restriction upstream condition, i.e., the downstream disturbance
will not change the flow rate across the choke. By contrast, the flow behavior depends on both upstream
and downstream conditions of the restriction under the sub-critical condition. They also summarize some
typical empirical and theoretical models to calculate the flow rate at the respective regime. Equation 2.4
was documented by Al-Safran and Brill (2017) for single-phase liquid flow through chokes under the
sub-critical condition, which is similar to Equation 2.2 except for the restriction ratio not explicitly shown
in their equation. C depends on the type of restriction, the ratio of the restriction diameter to the
D
upstream pipe diameter (γ), and the Reynolds number with reference to the restriction diameter. They
also mentioned the equation for multi-phase flow under sub-critical flow conditions as Equation 2.5
developed by Beggs et al. (1980), which was the same as the one for single-phase liquid flow but with
density and velocity replaced by those of the mixture.
(cid:115)
2∆P
Q=C A (2.4)
D o ρ
ρ V2
∆P = m mo (2.5)
2C2
D
2.2.2 Studies of Fluid Flow Downstream of Restrictions
Choking or restriction has profound effects on downstream flow behaviors. It can change the flow
pattern due to excessive mixing, influence downstream pressure drop, induce dispersion, and further impact
the drop size. The following reviews studies related to fluid flow downstream of inlet choking, with a focus
on flow pattern, pressure gradient, and droplet size.
2.2.2.1 Flow Pattern
Choking can significantly impact the flow pattern downstream in a pipe. Shmueli et al. (2018) observed
that the oil-water flow pattern changed from stratified flow to dispersed flow for a 0.2 bar pressure drop
choking implemented at the inlet. This alternation was also verified by measured vertical water cut profiles
of the pipe cross-section at a downstream distance of 120 pipe diameters, which demonstrated a
homogeneous water cut profile for the 0.2 bar inlet mixing compared to a stratification of the water cut
profile without inlet choking.
20
|
Colorado School of Mines
|
2.2.2.2 Pressure Drop
This part reviews studies related to inlet choking and its effect on downstream pressure gradient, and
some other factors including pipe diameter, mixture velocity, water cut, and oil viscosity are also reviewed.
Effect of choking level
Schu¨mann et al. (2016) used an inlet static mixer to create dispersed flow in a 25 m long horizontal.
They compared the measured pressure drop at a distance of 200 pipe diameters downstream of the inlet
mixer for the cases with and without the mixer and concluded that there existed a substantial pressure
increase for the water-dominated flow with the presence of a mixer. By contrast, the pressure drop almost
stayed constant for the oil-dominated flow with and without the mixer (Figure 2.18).
Figure 2.18 Comparison of downstream pressure gradient with and without inlet choking (Schu¨mann et al.,
2016).
Effect of the pipe diameter
A smaller pipe diameter induces a larger wall shear stress and thus higher pressure drop at a given flow
condition. Zhang and Xu (2016) studied emulsion flow in 25-mm. and 50-mm. inner diameter horizontal
pipes with a length of 2 meters. White mineral oil and tap water were used as the testing fluids. A special
static mixer was installed at the pipeline inlet section to facilitate the emulsion creation. The measured
wall shear stress as well as the pressure drop are larger for the smaller pipe, as shown in Figure 2.19 and
Figure 2.20. It can also be concluded that the pressure drop is more sensitive to mixture velocity in the
smaller diameter pipe.
21
|
Colorado School of Mines
|
Figure 2.21 Pressure drop as a function of the oil volume fraction in emulsion pipe flow (Zhang and Xu,
2016).
Effect of water cut
Zhang and Xu (2016) created oil-water emulsions in the pipe with an inlet mixer and they mentioned
that the dispersion was pretty good without noticing any evident droplet stratification. The pressure drop
first increased with increasing the water cut before the inversion point and then decreased with a further
water cut increment after the inversion point. The pressure drop changed more rapidly for a lower water
cut than that for a higher water cut after the inversion. In particular, the observed phase inversion was
invariably at an 80% oil volume fraction regardless of the pipe diameter and the mixture velocity
(Figure 2.21).
Shmueli et al. (2018) examined the effect of water cut on the downstream pressure drop of oil-water
flow through a ball valve in a 69 mm inner diameter horizontal pipe with a length of 51 m. When no
choking was applied, they observed stratified flow at a mixture velocity of 0.5 m/s, stratified flow with a
mixing interface at 1.0 and 1.5 m/s, and dual dispersion for a mixture velocity of 2.0 m/s. However, the
flow patterns all changed to dispersed flow when a choking level of a 0.2 bar pressure drop was applied at
the inlet. The pressure gradient without inlet choking demonstrated a smooth increase with increasing the
water cut. Such an increase was believed to be caused by the formation of a full dispersion of the oil phase,
which was a densely packed oil droplet layer of high viscosity in the upper region of the pipe. The pressure
gradient variation with the water cut for the 0.2 bar inlet choking was related to the presence of phase
inversion. The pressure gradient increased with increasing water cut till 50%, then decreased with a further
increase thereafter. However, the pressure gradient measuring position was not reported in this study and
only one choking level was tested and compared.
23
|
Colorado School of Mines
|
Figure 2.22 Pressure gradient 200 L/D downstream of the valve at different mixture velocities (Schu¨mann
et al., 2016).
Effect of oil viscosity
Schu¨mann et al. (2016) measured pressure drop downstream at a distance of 200 pipe diameters with
oil of different viscosities and found that pressure gradient increased with the oil viscosity for oil continuous
flow, while the water continuous flow did not show a dependency in the pressure gradient for different oil
viscosities (Figure 2.23).
2.2.2.3 Droplet Size
Extensive studies have been conducted to investigate the effect of choking levels on downstream droplet
size in oil-water two-phase flow. Furthermore, the effect of other factors including development length,
mixture velocity, and water cut on droplet size have also been reviewed.
Effect of choking level
Pressure drop across the valve is a major droplet breakup mechanism. A higher pressure drop is
believed to induce a smaller droplet size shown by Figure 2.24 (Fossen and Schu¨mann, 2017). The DSD
significantly shifts towards the left, which indicates a smaller droplet size and a narrower distribution, as
shown in Figure 2.25. The authors used a 100 mm inner diameter pipe with a butterfly valve to study the
relative effects of flow rate and pressure drop across a valve on the droplet breakup process. A key finding
is the dominant role of pressure drop across the valve compared to flow rate and water cut. It is worth
mentioning that only droplet size variations at 1 m downstream of the valve under different flow conditions
were studied in their paper.
24
|
Colorado School of Mines
|
Figure 2.25 Comparison of the DSD as a function of pressure drop across the valve for different water cuts
at a flow rate of 20 m3/h (Fossen and Schu¨mann, 2017).
Paolinelli et al. (2018) measured both the volume density and cumulative volume droplet size
distribution at different choking levels. They found that a higher choking level caused a smaller droplet size
downstream with a narrower volume density distribution and a steeper increase in cumulative droplet size
distribution (Figure 2.26).
Figure 2.26 Volume density droplet size distribution for various mean energy dissipation rates at a 5%
water cut (Paolinelli et al., 2018).
Skjefstad et al. (2020) investigated the inlet choking level on the separation performance of a new
separator. They measured the droplet size at the position 1 m downstream of a ball valve in a 67.8 mm
inner diameter horizontal pipeline with a length of 13 m. They observed a droplet size decrease with
increasing choking levels. Interestingly, the case without a choke gave a much smaller droplet size. They
26
|
Colorado School of Mines
|
pointed out that most of the oil and water existed as continuous phases without choking, leading to very
small droplets entrained in the liquid. Those small droplets were not separated out during the circulation.
They believed that this was why smaller droplet sizes were observed when no choke was applied. In fact, it
is also related to the flow pattern and droplet sampling position. In Figure 2.27, the flow pattern is dual
dispersion without choking and oil-in-water dispersion with choking. The sampling position is at the
bottom indicating oil droplets are sampled for all cases. The dual dispersion case can have fewer entrained
droplets and oil droplets tend to migrate upward to the interface, which makes the bottom sampling
location give a much smaller droplet size.
Figure 2.27 Cumulative volume fraction at the respective droplet size at different choking levels measured
at the bottom of the pipe (mixture velocity is 500 L/min with water cut 50%) (Skjefstad et al., 2020).
Effect of development length
Paolinelli et al. (2018) investigated the effect of the development length on the droplet size evolution
downstream of a globe valve in a 0.1 m inner diameter horizontal pipe. The droplet size was measured at
both 2.1 m (21 pipe diameters) and 6.1 m (61 pipe diameters) downstream of the valve. The authors found
that the maximum droplet size (d and d ) did not vary much at those two positions. Nevertheless,
max 98
Sauter mean diameter (d ) showed an increase downstream. The authors believed this increment was due
32
to the coalescence of medium size droplets which had a high frequency. However, the mean droplet size
growth did not exceed 9%, which made the authors decide to monitor the droplet size only at 2.1 m
downstream of the valve (Figure 2.28).
27
|
Colorado School of Mines
|
Figure 2.28 Measured droplet sizes as a function of the distance downstream the valve at different pressure
drop across the valve (Paolinelli et al., 2018).
Voulgaropoulos and Angeli (2017) investigated oil-water two-phase dispersed flow created by an inline
mixer in a 26 mm inner diameter 4-meter-long horizontal pipeline. They used a combined method using
high-speed planar laser-induced fluorescence and particle image velocimetry to obtain droplet size
information. Silicone oil and 52% w/w water/glycerol were used as the oil and aqueous phases. The
mixture velocity ranged from 0.2 to 0.8 m/s. They measured the droplet size at two streamwise positions,
i.e., one is 15 pipe diameters downstream of the inline mixer, and the other one is 135 pipe diameters
downstream of the inlet mixer.
They measured the droplet concentration variation along the pipe and observed a phase stratification
downstream of the inlet mixer. Figure 2.29 compares the droplet concentration variation of O/W and W/O
dispersed flow with the same dispersed volume fraction downstream of the inlet mixer at two different axial
positions (15 L/D and 135 L/D). For O/W dispersed flow, the oil droplets tend to migrate to the top area
of the pipe cross-section, while water droplets settled down to the bottom of the pipe for W/O dispersed
flow. Big water droplet coalescence occurred downstream which induced a free water layer at the pipe
bottom. Notably, a clear water layer developed in all cases at the axial position of 135 pipe diameters.
They also pointed out that drop migration was mainly caused by gravity, lift forces, and shear-induced
diffusion. In addition, they concluded that the dispersed oil droplet size was much smaller than that of the
dispersed water droplet. They explained that this was because the oil viscosity (0.0046 Pa·s) is smaller
than the water mixture (a 52% w/w water/glycerol) viscosity (0.0084 Pa·s). The energy dissipation in the
mixer is lower when oil is the continuous phase due to a smaller friction factor.
28
|
Colorado School of Mines
|
Figure 2.29 Planar laser-induced fluorescence images at the two axial measuring locations for two flow
conditions with the scale bar of 5 mm long on the bottom image. ϕ is the input oil fraction
o
(Voulgaropoulos and Angeli, 2017).
Effect of water cut
Paolinelli et al. (2018) compared the effect of different water cuts on the droplet size downstream of a
globe valve. They found that a higher water cut generally induced a broader droplet size distribution. The
cumulative droplet size distribution variation was not pronounced but seemed to demonstrate a steeper
change for the lower water cut case (Figure 2.30).
The literature review of this part shows that studies concerning fluid flow downstream of restrictions
mainly focus on two-phase oil-water flow and lack of deep understanding. Many researchers used an inlet
choke without systematically examining its effect on downstream flow behaviors.
2.3 Previous Modeling Studies on Two-phase Slug Flow
Gas-liquid two-phase slug flow models were used by several researchers for modeling three-phase slug
flow by treating the oil-water phase in gas-oil-water three-phase flow as one phase (Hall, 1992 and
Dehkordi et al., 2019b). In this case, the gas-oil-water three-phase flow could be simplified to gas-liquid
two-phase flow. Three major two-phase slug flow models are reviewed in this section, including Dukler and
Hubbard (1975) model, Taitel and Barnea (1990) Unified model, and Zhang et al. (2003) Unified model.
29
|
Colorado School of Mines
|
Figure 2.30 Volume density droplet size distribution at different water cuts (Paolinelli et al., 2018).
2.3.1 Dukler and Hubbard (1975) Model
Dukler and Hubbard (1975) proposed a model to predict the pressure drop across the slug body of
gas-liquid slug flow in horizontal pipes. One major assumption is that the pressure drop in the gas pocket
of the film region is negligible. Therefore, they stated that the pressure drop across a liquid slug body
consisted of two parts as shown by Equation 2.6: pressure drop induced by the acceleration of the liquid
film to the velocity of the slug (Equation 2.7), and pressure drop caused by the wall friction to the slug
(Equation 2.8), as shown in Figure 2.31. Another key assumption implicitly made by the authors is that
the film velocity and the film holdup behind a slug were a function of the position from the rear of the slug
due to the deceleration in the film after the slug. They believed the film holdup decreased rapidly to a
value close to H right behind the slug and further declined only slightly. H denotes the liquid
LFE LFE
holdup in the film region right in front of the slug. Similarly, V is the average velocity of the liquid film
FE
just in front of a slug.
∆p =∆p +∆p (2.6)
s a f
x
∆p = (V −V ) (2.7)
a A S FE
30
|
Colorado School of Mines
|
Figure 2.31 The physical model for horizontal slug flow modified from Dukler and Hubbard (1975).
f [ρ H +ρ (1−H )]V2(l −l )
∆p = S L LS G LS S S m (2.8)
f 2D
To obtain the acceleration pressure drop ∆p , the authors proposed an equation to calculate the mass
a
rate of pickup (x), as shown by Equation 2.9. They also proposed Equation 2.10 for the translational
velocity and suggested using C =0.021ln(Re )+0.022 for approximating the constant C when Re has
T S T S
a range of 30000-400000.
x=ρ AH (V −V ) (2.9)
L LFE T FE
V =(1+C )V (2.10)
T T S
They derived another important equation to relate the average liquid film velocity V with the liquid
F
holdup in the film region based on liquid film material balance, given in Equation 2.11. Note H and V
LF F
are not constant values but change with position along the flow direction in the film region.
H −H
V =V [1−C ( LS LF)] (2.11)
F S T H
LF
To find the pressure drop of wall friction, the authors introduced an equation (Equation 2.12) to get the
length of the mixing eddy (l ) where s was the specific weight of the liquid phase.
m L
0.3ρ (V −V )2
l = L S FE (2.12)
m 2s
L
The friction factor can be found by Equation 2.13 and Equation 2.14 (Hall, 1992).
31
|
Colorado School of Mines
|
0.125
f =0.00140+ (2.13)
S Re0.32
S
ρ H +ρ (1−H )
Re =DV L LS G LS (2.14)
S Sµ H +µ (1−H )
L LS G LS
Dukler and Hubbard (1975) derived an equation to get the slug length based on a material balance of
the liquid in the film region, given in Equation 2.15. The parameter H in Equation 2.15 was obtained
LFE
by the hydrodynamic of the liquid film based on the momentum conservation, Equation 2.16 and Equation
2.17.
V V
l = S [ SL −H +C (H −H )] (2.15)
S ω(H −H ) V LFE T LS LFE
LS LFE S
(cid:90) HLFS l
f(H )dH = F (2.16)
LF LF D
HLFE
C T2H L2 S − 1 [π 2sinα 2HLF−sin2α 2 −0.5cosα]
f(H )=
H L2
F
Fr 1−cosα 2
(2.17)
LF f FB2α
π
+sinβH FL rF
Where B =1−C T(HLS H− LH FLF)), H
LF
= α− 2s πinα, F
r
= gV DS2
Nicholson et al. (1975) modified the Dukler and Hubbard (1975) model. In the original model, the
authors assumed the film solution started with the conditions of the slug H =H . However,
LFS LS
Nicholson et al. (1975) found the existence of a critical mixture velocity V∗, below which the numerator of
m
Equation 2.17 became negative. In this case, H needed to be adjusted downward from H until
LFS LS
f(H ) first becoming positive. Then, the film solution should start with this adjusted value instead of
LF
H . When V >V∗, the starting condition of H =H was suggested to be used. The authors
LS m m LFS LS
pointed out that V∗ was proportional to pipe diameter at a given fluid system.
m
Another difference from the original model is the equation for the slug length. The authors derived
Equation 2.18 assuming that V H over the film region was well approximated by its end condition
F LF
V H .
FE LFE
l (V H −V )V
l = F FE LFE SL S (2.18)
S V −H (V +V )
SL LS SL SG
32
|
Colorado School of Mines
|
2.3.2 Taitel and Barnea (1990) Unified Model
Taitel and Barnea (1990) are the first to propose a unified model applicable to all inclination angles for
two-phase slug flow. The liquid entrainment in the gas core of the film region is not considered. They used
a slug unit consisting of a slug zone and a film zone for modeling two-phase slug flow (Figure 2.32). They
introduced three cases based on various degrees of simplicity for the hydrodynamics of the liquid film. Our
main focus is Case 3 from their original work, which assumes a constant film thickness. The key steps in
solving the problem are summarized in the following.
Figure 2.32 Two-phase slug flow model (Taitel and Barnea, 1990).
The average liquid velocity in the slug (u ) is solved by Equation 2.19.
L
V =V +V =u H +u (1−H ) (2.19)
s SL SG L LS b LS
Guess a liquid film thickness and calculate all the wetted parameters.
The liquid velocity in the film (u ) is solved with Equation 2.20 and the gas velocity in the film region
f
(u ) is solved with Equation 2.21. Then, all the friction factors and shear stresses are calculated.
G
(u −u )ρ AH =(u −u )ρ AH (2.20)
t L L LS t f L LF
V =u (1−H )+u H (2.21)
s G LF f LF
33
|
Colorado School of Mines
|
The momentum equation (Equation 2.22) is checked to see if it is converged. If not, go back to the step
of guessing a new liquid film thickness till the momentum equation converges.
τ S τ S 1 1
f f − G G −τ S ( + )+(ρ −ρ )gsinβ =0 (2.22)
A A I I A A L G
f G f G
The film length (l ) is calculated by Equation 2.23.
F
l (V −u H )=l (u H −u H ) (2.23)
U SL f LF S L LS f LF
Eventually, the pressure drop of the slug unit can be calculated by Equation 2.24.
τ πDl (τ S +τ S )l
∆P =ρ gsinβl + S S + f f G G F (2.24)
u u U A A
It is noteworthy to mentioning that the authors suggested to use slug length as a model input rather
than the slug frequency, though those two were used interchangeably. They explained the reason was that
the slug length was based on a physical model, while the slug frequency was typically obtained by
experiments.
2.3.3 Zhang et al. (2003) Unified Model
Zhang et al. (2003) used the film region as the control volume to derive a unified model encompassing
all inclination angles. They also argue that slug flow has transition boundaries to several other flow
patterns, therefore their unified model can also be used to predict other flow patterns. The continuity
equations, momentum equations, closure relationships, and solution procedure are introduced below.
Continuity equations: considering the mass flow balance in the control volume, i.e., mass flowing into
the control volume equals the mass flowing out, continuity equations for liquid and gas phases could be
written as Equation 2.25 and Equation 2.26, respectively.
H (V −V )=H (V −V )+H (V −V ) (2.25)
LS T S LF T F LC T G
(1−H )(V −V )=(1−H −H )(V −V ) (2.26)
LS T S LC LF T G
Two other continuity equations were also given by the authors, as shown by Equation 2.27 and
Equation 2.28.
l V =l H V +l (H V +H V ) (2.27)
U SL S LS S F LF F LC G
34
|
Colorado School of Mines
|
l V =l (1−H )V +l (1−H −H )V (2.28)
U SG S LS S F LF LC G
Momentum equations: based on the force balance, the momentum equations for the liquid film and gas
pocket could be written as Equation 2.29 and Equation 2.30, respectively. A combined momentum equation
Equation 2.31 could be obtained by eliminating the pressure drop from Equation 2.29 and Equation 2.30.
ρ (V −V )(V −V ) τ S −τ S p −p
L T F S F + I I F F −ρ gsinβ = 2 1 (2.29)
l H A L l
F LF F
ρ (V −V )(V −V ) τ S +τ S (p −p )
GC T G S G − I I G G −ρ gsinβ = 2 1 (2.30)
l (1−H )A G l
F LF F
ρ (V −V )(V −V )−ρ (V −V )(V −V )
GC T G S G L T F S F =
l
F (2.31)
(τ S −τ S ) (τ S +τ S )
I I F F + I I G G −(ρ −ρ )gsinβ
H A (1−H )A L G
LF LF
Closure relationships:
Liquid entrainment
Liquid entrainment in the gas core was defined as Equation 2.32 for slug flow (Zhang et al., 2003).
They also used Oliemans et al. (1986) equation in its dimensionless form to obtain the liquid entrainment
in the gas pocket (Equation 2.33).
H V
f = LC G (2.32)
E H V +H V
LF F LC G
f
E =0.003We1.8Fr−0.92Re−1.24Re0.7(ρ /ρ )0.38(µ /µ )0.97 (2.33)
1−f SG SG SG SL L G L G
E
Where We SG = ρGV σS2 GD, Fr SG = √VS gG D,Re SL = ρLV µS LLD, Re SG = ρGV µS GGD
Shear stresses
The wall shear stresses and the interfacial shear stress were obtained by Equation 2.34 and Equation
2.35. f and f were calculated by Equation 2.36 and φ denoted phases and can be F and G. C and n are
F G f
16 and 1 for laminar flow or 0.046 and 0.2 for turbulent flow.
ρ V2 ρ V2
τ =f L F,τ =f G G (2.34)
F F 2 G G 2
35
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.