Oil Droplet Size Distributions in Deep-Sea Blowouts: Influence of

May 15, 2018 - ... of Technology, Eißendorfer Straße 38, 21073 Hamburg , Germany ..... ηl is dynamic viscosity of the oil, ul is exit velocity at t...
4 downloads 0 Views 4MB Size
Download PDF
Article Cite This: Environ. Sci. Technol. XXXX, XXX, XXX−XXX

pubs.acs.org/est

Oil Droplet Size Distributions in Deep-Sea Blowouts: Influence of Pressure and Dissolved Gases Karen Malone,*,† Simeon Pesch,‡ Michael Schlüter,‡ and Dieter Krause† †

Institute of Product Development and Mechanical Engineering Design, Hamburg University of Technology, Denickestraße 17, 21073 Hamburg, Germany ‡ Institute of Multiphase Flows, Hamburg University of Technology, Eißendorfer Straße 38, 21073 Hamburg, Germany S Supporting Information *

ABSTRACT: To date, experimental investigations to determine the droplet size distribution (DSD) of subsea oil spills were mostly conducted at surface conditions, i.e. at atmospheric pressure, and with dead, i.e. purely liquid, oils. To investigate the influence of high hydrostatic pressure and of gases dissolved in the oil on the DSD, experiments with a downscaled blowout are conducted in a high-pressure autoclave at 150 bar hydrostatic pressure. Jets of “live”, i.e. methane-saturated, crude oil and n-decane are compared to jets of “dead” hydrocarbon liquids in artificial seawater. Experiments show that methane dissolved in the liquid oil increases the volume median droplet diameter significantly by up to 97%. These results are not in good accordance with state-of-the-art drop formation models, which are based on oil-only experiments at atmospheric pressure, and therefore show the need for a modification of such models which incorporates effects of hydrostatic pressure and dissolved gases for the modeling of deep-sea oil spills and blowouts.

1. INTRODUCTION The oil well blowout causing the explosion on the drilling platform Deepwater Horizon in April 2010 in the Gulf of Mexico was unprecedented in its extent and in the water depth in which it took place. For 87 days, about 5 million barrels of oil and 170 000 t of gas (C1−C5) spilled into the ocean in a depth of approximately 1500 m.1,2 Because of the deep-sea nature of the spill, up to 50% of this oil1 stayed below the sea surface as a result of density stratification and dispersion. To numerically model the mass distribution, propagation, and biodegradation of the oil in the ocean for such a deep-sea blowout, the initial oil droplet size distribution (DSD) is one of the key parameters.3−12 For a long time, models for the prediction of droplet size distributions were mainly based on experiments in stirred tank reactors. For example, Boxall et al. proposed a calculation of water-in-oil droplet sizes based on the turbulent kinetic energy for a stirred tank system.13 Since the 1990s and especially since the Deepwater Horizon blowout, different approaches to determine the droplet size distribution of oil-in-water jets have been developed. Various research groups conducted experiments with varying nozzle diameters, volume flow rates, and oil types at atmospheric or slightly increased pressures.14−17 A single large-scale deep-sea field experiment (“DeepSpill”) was performed off the coast of Norway in 200018 using oil-only and combined oil−gas releases where the phases were mixed just prior to injection, not allowing for © XXXX American Chemical Society

dissolution of the gas in the oil. On the basis of and validated against these experiments, several numerical models for droplet formation have been developed. The VDROP-J model by Zhao et al. uses a population-balance approach including breakup and coalescence of droplets based on the turbulent kinetic energy in a jet to predict the occurring size distribution.12,14,19 Many current models for near-field oil propagation use the approach of Johansen et al.,20 who suggested a modified Weber number scaling based on small-scale, oil-only experiments and DeepSpill.6 Li et al.21 proposed a Unified Droplet Size Model which correlates the volume median diameter to a combination of the Weber number and Ohnesorge number. But while there is today a sound understanding of drop formation for liquid−liquid jets, the influence of high hydrostatic pressure and dissolved gases on drop formation processes has hardly been investigated to date. Because of the weak compressibility of liquids, hydrostatic pressure is generally assumed to have a negligible effect on the DSD of oil-in-water emulsions. On the other hand, the physical properties of a gassaturated oil are severely affected by a high hydrostatic pressure due to pressure-dependent partitioning coefficients (e.g., refs 22−25), which will also alter its drop formation behavior. Aman Received: January 30, 2018 Revised: April 19, 2018 Accepted: May 4, 2018

A

DOI: 10.1021/acs.est.8b00587 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology Table 1. Experimental Sets and Conditions case

a

liquid

dissolved CH4 (g/L liquid)a

1

LSC

0

2

LSC

3931

3

n-decane

0

4

n-decane

7225,32

run

liquid volume flow rate (L/min)

1.1 1.2 2.1 2.2 3.1 3.2 4.1 4.2

2.06 2.06 1.99 2.10 1.12 1.22 1.11 1.24

± ± ± ± ± ± ± ±

0.12 0.12 0.12 0.12 0.12 0.12 0.12 0.12

pressure pa (bar) 149.6 150.0 150.4 149.4 149.9 149.5 149.0 150.4

± ± ± ± ± ± ± ±

0.6 0.6 0.6 0.6 0.6 0.6 0.6 0.6

temperature (°C) 19.6 19.8 19.8 20.2 20.7 20.6 19.4 20.0

± ± ± ± ± ± ± ±

1 1 1 1 1 1 1 1

Estimated based on measurement data of the saturation concentration from literature.25,31,32

component liquid and also to widen the range of flow conditions due to different physical properties. As methane is usually the most abundant gaseous component in reservoir fluids (e.g., ≈ 80 mol % for DWH spill2), it was chosen to investigate the effect of dissolved gas on the DSD. Methane 3.5 (CH4; ≥ 99.95% pure) was used to presaturate the liquids for the “live oil” cases (2) and (4). All experiments were conducted at a hydrostatic pressure of 150 bar (autoclave and oil reservoir) and a temperature of 20 °C. Liquid volume flow rate was kept constant for each liquid, but reduced for n-decane compared to crude oil in order to cover a larger range of flow conditions. Oil releases for each experiment lasted for approximately 8.5 s (case 1 and 2) or 15 s (case 3 and 4). Droplet size measurements were performed over a period of 2.5 to 5 s after stabilization of the jet. Experimental sets and conditions are summarized in Table 1. 3.1. General Setup. The experimental setup is depicted in Figure 1. Experiments were conducted within an acrylic cylinder (190-mm inner diameter, 600-mm height) filled with artificial seawater. The seawater (salinity 3.5%) was prepared according to Kester et al.33 The cylinder was placed within the high-pressure steel autoclave which was filled and pressurized with tap water. A flexible membrane connected to the acrylic cylinder ensured pressure equalization and therefore isobaric conditions between test volume (within the acrylic cylinder) and autoclave. The oil entered into the test volume via a 1.5mm nozzle (straight tube of 40-mm length upstream of the orifice) from a pressurized reservoir positioned outside the autoclave. The reservoir consisted of a stainless steel cylinder divided into two chambers by a freely moving brass piston; the oil was always placed in the lower chamber. The upper chamber was filled and pressurized with tap water. An equal-volume cylinder was connected to both the autoclave and the upper chamber of the reservoir. A linear drive with position tracking at the equal-volume cylinder was installed to generate an isobaric volume flow from the reservoir into the autoclave. A more detailed description of the functionality of this closedloop system can be found in Seemann et al.34 A ball-valve at the inflow pipe (valve V1 in Figure 1) was kept closed prior to the experiment in order to prevent premature mixing and was opened simultaneously to starting the linear drive of the equalvolume-cylinder. The temperature inside the autoclave was monitored using four Pt1000 resistors; a circulation pump at the bottom of the autoclave ensured a constant temperature over the height of the vessel. Pressure inside the autoclave (pa, equal to pressure inside the test volume), in the upper chamber of the oil reservoir (preservoir), and in the inflow pipe (ppipe) was monitored with pressure transmitters Keller PA-33X (pa, preservoir; pressure range 0 to 550 bar, precision ±0.55 bar) and Barksdale UPA 3

et al. were the first to use a pressurized, methane-saturated liquid for droplet size determination in a stirred autoclave cell,26 but as the technical specifications of the apparatus do not allow for a nonsaturated baseline, the effect of high pressure and dissolved gas cannot be quantified. Recently, Brandvik et al. reported droplet size distributions of methane-saturated crude oil under high pressure.27,28 As they reported the formation of numerous gas bubbles alongside the liquid oil, one has to assume that the fluid experienced a phase-change during the experiment, which by itself will affect the drop formation processes29 and therefore does not allow for a quantification of the effect of gas saturation on the DSD. To quantify the effects of dissolved gases on the droplet size distribution, experiments with downscaled hydrocarbon blowouts under artificial deep-sea conditions (150 bar hydrostatic pressure) were performed in a laboratory environment based on a test rig by Gust et al.30 Jets of purely liquid oils were compared to jets where the oil has been previously saturated with methane. This setup is analogous to the concept of “live” and “dead” oil as presented by Ahmed where “live oil” describes a crude oil as it exists in the reservoir, i.e. saturated with large amounts of dissolved C1 to C5.22 Compared to that, a “dead oil” describes an oil at atmospheric conditions which contains just very little or no C1−C5 components. In the following, the terms “live” and “dead” oil are used for the methane-saturated and nonsaturated liquids used in the experiments, respectively. Droplet size distribution for both “live” and “dead” oil jets are presented and compared to the predictions of the Weberscaling approach by Johansen et al. and the Unified Droplet Size Model by Li et al.

2. MATERIALS AND METHODS Droplet size distributions in subsea oil jets were measured in an adapted high-pressure test rig within a 99-L steel autoclave which was applied to generate the deep-sea conditions for the experiments. To carry out the experiments, an existing test rig was transformed from a mobile, mostly manually operated lab placed inside a 20-ft. container30 into a stationary research facility with fully automated control systems (see Supporting Information S-1). To investigate the effect of dissolved gas on the droplet size distribution, four different sets of experiments were carried out. Those consisted of (1) “dead” and (2) “live” crude oil and (3) “dead” and (4) “live” n-decane. Each set consisted of two individual experiments. Louisiana Sweet Crude oil (LSC) provided by BP (Marlin Platform Dorado, item ID A010E4), which was established as a surrogate oil for the investigation of the MC252/Deepwater Horizon spill, was used as crude oil. Ndecane (C10H22; ≥ 99% pure) was chosen to verify the findings from the crude oil data against a more chemically stable, singleB

DOI: 10.1021/acs.est.8b00587 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

pressure was released by removing water from the upper chamber and the system was allowed to equilibrate again. To enhance mixing of the fluids, the reservoir was tilted horizontally ±45° with a frequency of approximately 0.5 Hz to shorten the saturation time of the mixture from multiple days to 8−12 h. 3.3. Image Recording and Determination of Droplet Size Distribution. The emerging jet was recorded for surveillance from the side with a pressure-proof high-speed camera (Figure 2). Because of limited space inside the

Figure 2. Side view of a “live” crude oil jet (experiment 2.2). Red rectangle shows approximate size and position of the endoscope for droplet size analysis within the jet.

Figure 1. Test setup for jet experiments (not to scale).

(ppipe; pressure range 0 to 400 bar, precision ±0.4 bar), respectively. An additional differential pressure transmitter Wika DPT-10 (pressure range −10 to +10 bar, precision ±0.024 bar) was used to monitor the differential pressure between inflow pipe and autoclave (Δp = ppipe − pa). A HBM QuantumX MX840A universal amplifier was used to record the data at a sampling rate of 300 Hz. Liquid volume flow rates were calculated from the feed and piston diameter of the equalvolume cylinder. 3.2. Preparation of “Live Oil”. For experiments using “live oil”, the liquid hydrocarbons were saturated with methane prior to the experiments. For this purpose, the lower chamber of the reservoir was first filled with 300 mL of liquid (LSC oil or ndecane). The chamber was then pressurized with CH4. As data for the solubility of CH4 in either liquid at the exact experimental conditions are scarce, CH4 mass added was calculated to be at least twice the estimated saturation concentration of CH4 in the liquid at the target pressure.25,31,32 CH4 mass was measured with a SiTrans FC400 DN4 Coriolis mass-flow sensor with SiTrans Mass6000 transmitter; precision was ±0.35 kg/h. This larger amount of CH4 formed a gas layer on top of the oil, but as oil was removed from the bottom of the reservoir, it did not enter the experimental section. After methane was added to the oil, the reservoir was pressurized at the upper chamber with tap water up to 400 bar. All inlet valves were closed and the pressure in the reservoir was monitored until equilibrium, and therefore saturation, was reached.35 If the equilibrium pressure was above the target pressure of 150 bar,

autoclave, the camera was positioned vertically and equipped with a mirror tilted by 45°. For determination of the droplet size distribution, a high-pressure endoscope (Sopat VR-HP, Sopat GmbH, diameter 20 mm) was positioned concentrically to and 110 mm above the nozzle. This means a distance of 73 nozzle diameters from the exit, so the jet can be assumed fully dispersed at this position. Monochromatic images were recorded with an exposure time of 8 μs using incident light at a rate of 12 Hz (Figure 3). The endoscope’s measurement range of droplet diameters is 10 to 3000 μm; the focal plane was located at a distance of 500 μm (cases 1 and 2) and 1 mm (cases 3 and 4) from the lens. Droplets were detected and sized manually from the endoscopic images using the open-source software ImageJ.36 A minimum of 2000 droplets was evaluated for each experiment. Individual droplet diameters were determined (max error ±6.8 μm, combined from max magnification error and max manual sizing error) and grouped into logarithmically scaled bins. The median diameters of the number (dn50) and volume distribution (dv50) were determined for each experiment. The resulting empirical cumulative distribution function (CDF) of number was compared to a log-normal distribution function CDF(d) = C

1 1 ⎛ ln d − ln d i ⎞ ⎟ + ·erf⎜ 2 2 ⎝ 2 ·σ ⎠

(1)

DOI: 10.1021/acs.est.8b00587 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

(IFT) at the oil−water interface for “dead” and “live” LSC at the experimental conditions. Details on the measurement methods can be found in the corresponding data sets published via the Gulf of Mexico Research Initiative Information and Data Cooperative (GRIIDC).38−43 Material properties used in the calculations are summarized in Table 2. Literature data have Table 2. Material Properties at p = 150 bar, T = 20 °C Used in the Calculation of Reynolds and Weber Numbers case

liquid

1 2

LSC CH4-saturated LSC n-decane CH4-saturated n-decane

3 4

Figure 3. Top view from the center of a methane-saturated crude oil jet entering into seawater at 150 bar, 20 °C (experiment 2.2) for droplet size analysis.

d v50 = A ·We* − 3/5 D

(4)

where We =

D·ρl ·ul2

number, Re =

is the Weber number, Vi =

σl D·ρl ·ul ηl

We Re

54.6 ± 0.146 44 ± 447

3. RESULTS AND DISCUSSION The droplet size distribution was determined for oil jets of untreated (“dead”) and gas-saturated (“live”) LSC oil and ndecane under artificial deep-sea conditions (pambient = 150 bar). As reservoir and hydrostatic pressure were the same, no outgassing from the oil occurred and no free gaseous phase formed. The droplet sizes reported in the following therefore represent the entirety of the released fluid. The median diameters of number (dn50) and volume distribution (dv50) were evaluated for each experiment and the best fit of the experimental data to analytical cumulative distribution functions was determined. In comparison to different analytical distribution functions, the droplet size distribution of each experiment most closely follows a log-normal distribution function (R2= 0.83 to 0.994) according to eq 1, albeit with different spreading coefficients ranging from σ = 0.28 (case 1.2) to 0.72 (case 4.1). The experimental data for the cumulative distribution functions of number for cases 1−4 are depicted in Figure 4a−d alongside the best fit log-normal distribution. The Rosin−Rammler distribution function (eq 2) showed a significantly worse fit than a log-normal distribution, but might still be a reasonable correlation to the data (R2 = 0.81 to 0.95, see Figure 5 and Supporting Information S-2). A comparison of the “live oil” and “dead oil” cases for both liquids depicts a broader distribution for the “live oil” cases (2 and 4) than for the “dead oil” cases (1 and 3), respectively. n-Decane was distributed broader than LSC. The analytical distribution function for the determination

is viscosity

is Reynolds number, D is nozzle/vent

diameter, ρl is oil density, σl is oil−water interfacial tension (IFT), ηl is dynamic viscosity of the oil, ul is exit velocity at the nozzle/vent, and coefficients are A and B. In their original work, Johansen et al. suggested the empirical coefficients to be A = 15 and B = 0.8,20 later updating these to A = 24.8 and B = 0.08 using a larger data set.6 Those updated coefficients are used to compare the model with the experimental data in this work. For comparison with the Unified Droplet Size Model, the volume median diameter dv50 is also estimated using d v50 = r ·(1 + 10 ·Oh) p ·Weq D

1.08 ± 0.0545 0.385 ± 0.01945

22 ± 0.741 32 ± 143

been interpolated where the precise pressure and temperature conditions were not reported. Measurements of the interfacial tension (IFT) σ between LSC oil and artificial seawater show a large time dependence due to chemically impure interfaces.44 As droplet formation takes place immediately after the interface between the two phases is established, the IFT at t ≈ 0 s is applied for the calculation of the We number. Uncertainties in the density, viscosity, and IFT data as reported in Table 2 are taken into account for estimation of the dv50 according to eqs 3−5.

with di as median diameter, α and σ as spreading coefficients, ki = −ln(0.5) = 0.693, to determine the best fit to the empirical distribution using a χ2 test.37 To compare the results with the prediction by the model of Johansen et al.,20 the median volume diameter dv50 and modified Weber number We* were calculated for each experiment according to

(3)

741 ± 745 662 ± 0.723

15.7 ± 0.53940 24a 1.30+1.3 −0.65

oil−water IFT σl (mN/m)

As dynamic viscosity of CH4-saturated LSC has not been reported so far, data of the prespill downhole sample at the Macondo Well MC 252 (which also contained C2−C5, N2, H2S, and CO2) was used for calculation of Re. With regard to the different formulation of the liquid, this assumption was estimated to be accurate within a factor of 2.

(2)

⎡ ⎛ d ⎞1/3⎤ We* = We/⎢1 + B ·Vi·⎜ v50 ⎟ ⎥ ⎝ D ⎠ ⎥⎦ ⎢⎣

864 ± 1538 817 ± 1442

dynamic viscosity ηl (mPa*s)

a

and a standard Rosin−Rammler distribution function as suggested by Johansen et al.19 ⎛ ⎛ d ⎞α ⎞ CDF(d) = 1 − exp⎜⎜ −k i⎜ ⎟ ⎟⎟ ⎝ ⎝ di ⎠ ⎠

density ρl (g/ L)

(5)

We

where Oh = Re is the Ohnesorge number, r = 14.05, p = 0.460, and q = −0.518.21 Reynolds and Weber numbers were calculated for each experiment using material properties from experimental data based on literature and own measurements. Own measurements include (1) viscosity of “dead” LSC, (2) density of “live” and “dead” LSC, and (3) interfacial tension D

DOI: 10.1021/acs.est.8b00587 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

Figure 4. Cumulative droplet size distribution function of number: experimental data points and best fit by log-normal distribution function; legends show number median diameters dn50 and spreading coefficients σ for each experiment. (a) Case 1, pure LSC; (b) Case 2, methane-saturated LSC; (c) Case 3, pure n-decane; (d) Case 4, methane-saturated n-decane.

of the droplet size distribution from a given median diameter is controversially discussed in the literature: whereas Johansen et al. suggest a Rosin−Rammler distribution function for the volume distribution,20 Aman et al. report a good agreement of the experimental data to both a Rosin−Rammler and a lognormal distribution.26 Zhao et al. on the other hand find either a bimodal14 or a unimodal size distribution for different oils and discharge rates.14,48 They suggest that the bimodal distribution might be caused by incomplete breakup near to the discharge point.48 This argument also supports the assumption that the droplet sizes in our experiments had stabilized at the measurement position because the distribution is unimodal. Our results showed that the spreading factor not only varies widely between “dead” and “live” oil, but also between liquids. As the spreading factor determines the maximum and minimum diameter that will appear in an oil plume, it is a crucial factor in the correct prediction of near- and far-field oil propagation,

Figure 5. Comparison of experimental data, best-fit log-normal (spreading factor σ = 0.45), and best-fit Rosin−Rammler (spreading factor α = 2.5) distribution function of number for experiment 2.1.

Table 3. Results of Experiments 1.1−4.2: Reynolds Number and Weber Number, Median Diameters of Number (dn50) and Volume Distribution (dv50), and Minimum (dmin) and Maximum (dmax) Diameters case

liquid

1

LSC

2

CH4-saturated LSC

3

n-decane

4

CH4-saturated n-decane

run

Re

We

dn50 (μm)

dv50 (μm)

dmin (μm)

dmax (μm)

1.1 1.2 2.1 2.2 3.1 3.2 4.1 4.2

1607 1607 19714 17671 10889 11827 26974 30111

22235 22235 16768 13474 2279 2688 2468 3076

57 58 81 89 142 139 188 162

85 83 151 141 343 337 723 621

16 16 14 20 33 27 26 33

197 214 364 320 755 803 1319 1301

E

DOI: 10.1021/acs.est.8b00587 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology especially with regard to the amount of naturally dispersed oil. On the basis of the literature reviewed above and our own results, we believe that models for the prediction of the droplet size distribution of a subsea oil spill must include a possibility to predict an actual size distribution in addition to the median diameter. Measured droplet diameters ranged from a minimum of 16 μm (case 1) to a maximum of 1319 μm (case 4). Median diameters ranged from 57 μm (case 1) to 188 μm (case 4) for the dn50 and from 83 μm (case 1) to 723 μm (case 4) for the dv50. n-Decane (cases 3 and 4) produced altogether larger droplets than the Louisiana Sweet Crude oil (cases 1 and 2), which was to be expected due to the lower volume flow rate in those cases and significantly higher oil−water IFT for n-decane compared to LSC oil. Results of all experiments are summarized in Table 3. According to the experimental results, the presaturation of oil with methane can enhance the dv50 of a subsea oil spill by 74− 97% compared to the nonsaturated case with otherwise unchanged blowout conditions. Median droplet diameters of both number (dn50) and volume (dv50) distribution of the “live oil” cases 2 and 4 were significantly larger than those of the “dead oil” cases 1 and 3. To quantify the effect of dissolved methane on the median droplet diameter, the average dv50 of the “live” liquid is scaled by the average dv50 of the “dead” liquid for both crude oil and ndecane. For this purpose, a relational factor F=

Figure 6. Comparison between experimentally determined dv50 and estimation by models of Li et al.21 according to eq 5 and Johansen et al. according to eqs 3 and 4 (with A = 24.8 and B = 0.08).6 Vertical error bars indicate uncertainties of the diameter estimation due to uncertainties in material properties data (see Table 2); as these are