Professional Documents
Culture Documents
SKIMMER Oil Water Separator
SKIMMER Oil Water Separator
Oil/Water Separator
Joe Stires, * SPE, Conoco Inc.
Summary Discussion
At ultra-high water cuts (95% or more), the most dif- To take advantage of rising oil prices, many oil com-
ficult production problem may not be getting the oil out panies have increased production from older leases over
of the ground, but getting oil out of the water. While it is the past several years. In 1975, Continental Oil Co.
not uncommon to use large wash tanks to separate the oil (now Conoco Inc.) began a program to maximize pro-
from this type of flowstream, few designs can be tailored duction from the Big Lake field, Reagan County, TX.
to a specific application. This paper presents one such The field was the discovery field of the Permian Basin
design, including the equations used to analyze its per- region and has been on production since 1923. Produc-
formance. This type of analysis can be applied to any tion has been primarily from the Grayburg limestone, at
separator to predict what its performance should be. 3,000 ft (914 m). This reservoir has a strong water drive,
and water cuts currently average 90% or more. Liquid
Introduction productivities are usually more than 3000 BID (477
The increasing value of oil is making it economically at- m 3 I d) per well.
tractive to install high-volume artificial lift equipment in Most wells on our lease in this field were equipped
an oil well, and produce it to a WOR of 125 or more. with high-volume artificial lift in 1975-76 to take advan-
However, there are few methods for the design of an tage of the high productivities of these wells. The lease-
oil/water separator that will handle effectively liquid in- wide oil production increase from 290 to 525 BID (46 to
puts of 20,000 BID or more at these ultra-high water 83 m 3 I d) was accompanied by an 18,000- BID
cuts. By developing and presenting the design equations (2862-m 3 Id) increase in water. Total water production,
for one such design, this paper describes a foundation for 40,000 BID (6359 m 3 /d), exceeded the capacity of the
further work. treating facilities on this lease. As a result, oil carryover
Large-volume wash tanks often are used as oil/water into the produced-water disposal srstem increased, go-
separators or water polishers in high-rate, high-water-cut ing as high as 700 mg/L (0.7 g/dm ) in some tests. This
applications. Despite widespread use of wash tanks, meant losses of about 30 bbl (4.8 m 3 ) oil into the
there are no design equations available to aid in their disposal system daily and significant lost revenues. This
design 1. This paper presents several equations un separated oil also caused increased operating costs as
developed while adapting the API separator to oilfield water disposal wells began to plug.
use. It also evaluates design performance with field data. After reviewing these problems, we decided to build a
A comparison of actual performance with performance satellite battery to handle half the produced liquid on the
predicted by reaction kinetics equations shows that, for lease. The central battery was to be modified to increase
the most part, overall performance approached or ex- its treating and disposal capacity to 32,000 BID (5088
ceeded predicted performance. Although some flow m 3 I d) by installing some type of water-polishing device
characteristics failed to meet the modeled performance, between the existing free-water knockouts and the
we conclude that the design equations developed are produced-water disposal system.
valid for design of a high-water-cut, high-rate oil/water Although the capacity of this water-polishing device at
separator. ultra-high water cuts was the primary design criterion,
several other factors had to be considered.
• Now with Monsanto Oil Co. As a result, three secondary design criteria were
0149·2136/82/0010-8307$00.25
developed: that the vessel be built and installed quickly,
Copyright 1982 Society of Petroleum Engineers of AIME that it be reasonably priced, and that it be durable in the
BLIND
300 0
~
p= 0
0
0
-
!
300
"'0
o
100
Fig. 3-Performance of skim tank at 25,000 BID (3975 m 3/d).
1 2)
+ - d 1h2 +-d1hl cu fi/min,
expected.
To detennine the retention time of water in this vessel,
3 3 a tracer was mixed in water and injected into the
or flowstream before it entered the gas boot. Samples were
taken at the water outlet and then were analyzed for
tracer material. As indicated in Fig. 4, the peak retention
time was 19 minutes.
The degree of mixing is indicated by the width of the
retention curve at one half the peak height. In this test,
the width of the retention curve was 60 minutes using a
tracer concentration of 200 ppm. This indicates that tur-
bulence and mixing are occurring, but that they are not
affecting perfonnance adversely to any great extent.
Ct = r
o
Cdt
be plotted as a dimensionless curve.
To obtain a dimensionless plot, the ideal mean
residence time is used to reduce both t i and C i to func-
tions of dimensionless time, t D. Thus, for any sample,
or ti
tD=- .............................. (10)
00 7
.L:c X~ti ppm-min ....................... (5)
i and
o 7
E(tD)=-C i . . . • . . . . . . . . . . . . . . . . . . . . . . . . (11)
The retention time curve becomes the retention time Ct
distribution when Reaction engineering texts provide two simple single-
parameter models that represent two extremes of flow.
C These can be used to obtain clues to which flow regimes
E(t) = -minutes -I ....................... (6) are occurring in the vessel. These two equations also can
Ct be used to predict the retention-time distribution of this
is plotted vs. t. vessel, within the limiting assumptions.
The mean residence time, 7, of a vessel is a key The dispersion model assumes that flow is similar to
operating characteristic. If there is no dead space in the pipe flow, with mixing caused by turbulence and
vessel, molecular diffusion processes. As can be expected where
V one variable combines these two effects, the dispersion
7=-min, ............................... (7) coefficient, K, primarily reflects the amount of tur-
q bulence encountered during flow. The vessel dispersion
but the actual 7 for a vessel can be calculated from a number is K/(vL) , a dimensionless number used to
tracer test with the formula characterize the amount of mixing during flow. For
7="-0----
r Cxtdt
small values of K/(vL), [K/(vL)::;; 0.002], turbulence is
negligible and plug flow is described, resulting in a
Gaussian curve centered at t D = 1. As turbulence in-
r
creases, so does K/(vL), until fully mixed flow occurs as
k/(vL) approaches infinity. Increased mixing causes the
Cdt
distribution to reach a peak retention time before t D = 1
o and results in a lower peak concentration. For this
model,
00
.L:CXt~t
.....;O~--min . . ........................ (8)
Ct
7 is the first moment of the retention time distribution; ........................ (12)
the second moment is its variance, (12 (t) is
r
C(t-T)dt
(12t="-0 _ _-
The tanks-in-series model simply assumes flow
through a series of mixing tanks, each equal in size. The
single parameter describing this system is n, the number
r
o
Cdt
of tanks in this series. For this model,
E(tD)=
n(nt D)n-I e -ntD
(n-l)!
................... (13)
r
Ct 2dt
The second moment, the variance, of these functions is
2K K2
(12(tD)=- +8-- for the dispersion model . (14)
r
="-0-----_t 2
o
Cdt and
vL vL
~C·I Xt2·~t-72
L..J I
Note that as the number of tanks increases to infinity, or
= 0 as K gets very small, (12 (tD) approaches zero and overall
. . ................... (9)
flow through the system will approach plug flow.
Because the dispersion model allows for flow back
NOVEMBER 1982 2641
TABLE 2-COMPARISON OF Here it was assumed that n=3. From Eqs. 14 and 15, we
PROJECTED vs. ACTUAL PERFORMANCE AT 25,000 BWPD calculate that (J2(t D)=0.333, and KlvL=0.115. Be-
tween the two predictions is a composite curve. This is
Ideal Actual
Distribution an interpolation predicting what the retention time
Distribution
distribution should be for the constant-level skim tank
0.85 tD (46 min.) peak retention time 0.35 tD (19 min.)
0.78 tD (42 min.) mean retention time 1.01 t D (54 min.) based on the two models. According to this curve, the
0.86 tD (46 min.) 0.88 tD (47 min.) peak concentration is predicted to occur at t=0.85tD'
tD (54 min.) time at one-half peak 0.85 tD (46 min.) and the time span at one half the peak concentration is 7.
concentration The peak concentration should be about 0.9E(td)' The
0.9 peak concentration 0.99 variance, (J2, of this vessel, should be 0.333; the mean
residence time is 0.78tD' The actual retention time
distribution for this vessel is plotted in Fig. 6. It can be
seen that the peak concentration actually occurs at
t=0.35tD' From Eq. 8, 7=54.9 minutes (1.01tD) for
this distribution, and we can calculate that
RTD
1.0 (J2(tD)=0.303. Thus, in this vessel the actual constants
are n=3.29 and KlvL=0.106, using Eqs. 14 and 15.
0.8
The rapid increase of concentration in the actual
retention-time distribution seems to indicate the
0.6
-0
~
possibility of short circuiting in this vessel. However,
w
0.4
the mean residence time and the peak tracer concentra-
tion were nearly the same as predicted, and no slugging
0.2 was observed or recorded.
The retention-time distributions based on these con-
0 stants are plotted in Fig. 6. Comparing Figs. 5 and 6
0 05 1.5 2 shows how similar the actual performance was to the
ideal performance. The variance of the actual retention-
time distribution was only slightly less than the variance
Fig. 5-ldealized residence-time distributions at 25,000 BID
(3975 m 3 /d). predicted by the models. This is an indication that slight-
ly less mixing and turbulence was occurring than was
expected.
This conclusion is supported by two other com-
1.4 parisons. If the hump in the retention-time curve, which
begins at t= 1.2t D, is ignored as an anomaly, the time
1.2 during which measured tracer concentration exceeded
TANK IN one half of the peak concentration was O. 85t D, which is
1.0 SERIES
slightly less than the value expected. This indicates that
0.8 the actual flow regime was closer to plug flow than was
'0 predicted. The lack of turbulence also is indicated by the
~
UJ ACTUAL
0.6 DISTRIBUTION value of E(t D) achieved at peak concentration; this was
about 10% higher than was predicted by either flow
0.4 model (Table 2).
0.2
In a final comparison, the measured mean residence
time of 54.9 minutes occurred at 7= 1. 0 1t D, which is
0 slightly longer than the 7=0.78tD predicted by the
0 .5 1.5 2 model. In this distribution, samples taken after t=2t D
to could be regarded as coming from an inactive part of the
Fig. 6-Equivalent residence-time distributions at 25,000 vessel. If these data points are not considered, 7=47.6
BID (3975 m 3/d). minutes, or 0.88tD for this vessel, which agrees fairly
well with the predicted value.
According to Levenspiel,5 the difference between 7
for this smaller set of data, and the ideal 7 (= Vlq)
represents the amount of inactive volume in the vessel:
upstream, the distributions predicted by these two
models differ more and more with increasing deviation
from the plug flow regime. 5 Vactive
7 measured X
= ------
V
tideal
Analysis of Results and
Using these equations, the data from the retention time
t measured )
testing were used to arrive at a retention time distribution Vinactive =V - Vactive = V ( 1- .
for the vessel (Table 1). This distribution was plotted in tideal
Fig. 5 to predict what the retention time distribution for This vessel was designed to have a quiet space above
flow through the gas boot, suction piping, and spreader the uppermost spreader. In this vessel, 2 ft (0.61 m) of
areas of the constant-level skim tank should look like. space (or 14% of the working volume) was isolated, so
flowstream, ppm/min
d = diameter (see Fig. 2 for appropriate Gas boot: Design for surging to 48,000 BID (7631
subscripts), ft (m) m 3 /d) at u=0.5 ft/s (0.15 m/s), d=34 in. (86 cm),
E(t) = retention-time distribution of C with respect h=tank height+(2xd)+2 ft=24 ft (7.3 m).
to t, minutes - 1 Forebay riser: For velocity < 6 ftl s (1. 830 ml s),
E(t D) = retention-time distribution of C with respect d 1 :e:: 9.8 in. (25 cm).
to tD
Exit slots: For q=32,000 BID (5088 m 3 /d), use area of
h = height (see Fig. 2 for appropriate
100 sq in. (645 cm 2) or less to ensure velocity of at least
subscripts), ft (m) 3 ft/s (0.9 m/s). Assume eight slots [2 ftxO.5 in. (0.6
fi = average height mX 1.3 cm)] starting 1 ft (0.3 m) below top of spreader
K = effective (or longitudinal) dispersion coeffi- for both inlet and outlet slots.
cient, sq ft-min (m2-min) . Using a 21-ft and 16-inx 16-ft (6.6-mx4.9-m) tank
L = equivalent length of flow channel, ft (m) with 1,000-bbl (l59-m 3 ) nominal capacity, a spreader