Journal of Petroleum Science and Engineering 177 (2019) 496–507

Contents lists available at ScienceDirect

Journal of Petroleum Science and Engineering

journal homepage: www.elsevier.com/locate/petrol

Determination of binary diffusion coefficients between hot liquid solvents T

and bitumen with X-ray CT
Motonao Imaia,∗, Kazuaki Mikamia, Tatsuya Suganumab, Yoshihiro Tsuchiyaa,
Kazunori Nakagawab, Satoru Takahashia
a
Japan Oil, Gas and Metals National Corporation (JOGMEC), 1-2-2 Hamada, Mihama-ku, Chiba, 261-0025, Japan
b
Japan Petroleum Exploration Co., Ltd. (JAPEX), 1-2-1 Hamada, Mihama-ku, Chiba, 261-0025, Japan

A R T I C LE I N FO A B S T R A C T

Keywords: The Steam Assisted Gravity Drainage (SAGD) process has been used for in-situ thermal bitumen recovery in
SA-SAGD Canada. Nevertheless, there are several drawbacks that could impair the SAGD performance. One of the main
Bitumen drawbacks of SAGD is nontrivial consumption of energy to generate steam. To curtail the consumption of steam,
Solvent many SAGD variants have been proposed. For example, Vapor Extraction (VAPEX) was proposed as a solvent-
Diffusion
based cold production method and Solvent-Assisted SAGD (SA-SAGD) was proposed as a solvent-assisted hybrid
Boltzmann-Matano
method. In those solvent-based/assisted methods, the dissolution of solvents in bitumen enhances the in-situ oil
CT scan
mobility.
The speed of the mixing between solvents and bitumen in solvent-based/assisted recovery processes is dic-
tated by dispersive mass flux which is proportional to the transverse dispersion coefficient. Diffusion coefficient,
tortuosity and transverse dispersivity should be quantified to characterize the transverse dispersion coefficient.
In this research, the diffusion coefficients of solvents in Athabasca bitumen were determined with X-ray com-
puted tomography (CT).
A medical X-ray CT scanner was used to capture the diffusion phenomenon occurring between fresh liquid
solvents and bitumen. The experiments were designed to reproduce the in-situ high pressure, high temperature
condition (3.5 MPa, 135C) where solvent and bitumen actually interact at the vapor chamber edge in the SA-
SAGD process. Using our new °experimental system, we successfully measured molecular diffusion coefficients at
the condition of the vapor chamber edge for the first time.
In conclusion, the binary diffusion coefficients between solvents and bitumen were successfully measured at
the condition where the mixing of solvents and bitumen actually occurs in the SA-SAGD process. The char-
acterized diffusion coefficients will be input parameters for the reservoir simulation and semi-analytical models.

1. Introduction
Steam Assisted Gravity Drainage (SAGD) has been one of the most
prevailing in-situ thermal oil recovery techniques for decades in
Canadian bitumen recovery projects. In the SAGD process, driving force
is not pressure gradient but gravitational potential. As Butler (1998)
described in his book, oil flows downward along the interface between
expanding steam and existing bitumen. Since steam gives huge latent
heat to cold bitumen when steam is condensed to water, bitumen be-
comes movable along the edge of steam expanded area which is called
steam chamber. This induced oil flow is actually perpendicular to the
temperature gradient, which makes the SAGD process different from
other displacement processes where the direction of oil flow is basically
parallel to the pressure gradient. This kind of recovery process evoked
by gravity is called gravity drainage. To collect the dropped oil effi-
ciently, SAGD producers are drilled horizontally so as to lie on the re-
servoir bottom. Typically, injectors are drilled just a few meters above
on producers to be paired (Fig. 1).
Although SAGD has been widely applied in in-situ heavy oil re-
covery projects in Canada, problems emerge. One of the critical issues
on SAGD operation is the cost of steam generation (Ogino, 2004). Ty-
pically, steam is generated by gas-driven boilers. Since produced bi-
tumen is almost pure dead oil, fuel gas should be purchased from
outside resources. Efforts have been made to curtail steam consumption
or to enhance steam to oil ratio (SOR). In the late 80s, Vapor Extraction
(VAPEX) was proposed by Butler and Mokrys (1989). In VAPEX process,

∗
Corresponding author.
E-mail address: imai-motonao@jogmec.go.jp (M. Imai).

https://doi.org/10.1016/j.petrol.2019.02.049
Received 15 November 2018; Received in revised form 14 February 2019; Accepted 14 February 2019
Available online 20 February 2019
0920-4105/ © 2019 Elsevier B.V. All rights reserved.
M. Imai, et al.
Fig. 1. Schematic of SAGD process (by courtesy of Japan Petroleum Exploration
Co., Ltd.).
solvents which consist of hydrocarbon(s) are injected into SAGD wells
instead of steam. Solvent selection varies depending on operation
conditions but usually propane or butane have been used. Viscosity
reduction is achieved not by thermal effect but by solvent dissolution
into bitumen. Hence, heat conduction in SAGD corresponds to dis-
persive mass transfer in the VAPEX process. In other words, VAPEX is a
solvent analogue of SAGD. However Ardali et al. (2012) mentioned that
VAPEX faced difficulties when the industry attempted pilot tests and
field applications. Because of the slowness of dispersive mass transfer
compared with thermal conduction, recovery cannot be enough en-
hanced.
To overcome the limitation of VAPEX and its variant processes,
hybrid solvent-heat recovery processes have been proposed. Those
processes include Solvent-Aided Processes (SAP) (Gupta et al., 2005),
Solvent-Assisted SAGD (SA-SAGD) which is virtually same as the pro-
cess so called Expanding-Solvent SAGD (ES-SAGD) (Nasr et al., 2003),
Liquid Addition to Steam to Enhance Recovery (LASER) (Leaute, 2002)
and Steam Alternating Solvent (SAS) (Zhao, 2004).
According to Ardali et al. (2012), some Canadian E&P companies
have conducted field pilot tests in Encana at Christina Lake (SAP),
Imperial Oil at Cold Lake (LASER) and Nexen at Long Lake (ES-SAGD).
Among the past field test cases, the Christina Lake SAP where butane
was co-injected with steam into SAGD wells improved bitumen pro-
duction by 30% and SOR, which is regarded as one of the economic
indicators of the SAGD performance, by 25% (Bayestehparvin et al.,
2018); the Cold Lake LASER where the mixture of steam and pentane
plus (C5+) is injected and produced in huff-and-puff mode at a vertical
well marked 35% improvement of bitumen production along with
about 70% of solvent recovery (Ardali et al., 2012). The Long Lake ES-
SAGD also improved bitumen production rate by 6% and SOR by 7%
(Orr, 2009). In some cases such as Suncor at Firebag and Nexen at Long
Lake (both in ES-SAGD), no clear indication of improvement on bi-
tumen production and SOR was observed (Chen and Keshavarz, 2015;
Bayestehparvin et al., 2018). Obviously, the performance of solvent
assisted methods varies. It depends on various factors including diffu-
sion/dispersion characteristics, solvent selection, solvent fraction in
vapor, temperature and/or pressure conditions and reservoir hetero-
geneity.
In the context of SAGD-based solvent processes such as VAPEX and
SA-SAGD, the diffusive mass transfer of liquid solvents in bitumen-sa-
turated oil sands is the key to quantitatively characterize the process.
The speed of the mixing between solvents and bitumen in those pro-
cesses is dictated by mass transfer caused by the hydrodynamic trans-
verse dispersion (Bayestehparvin et al., 2018; Rabiei Faradonbeh et al.,
497
Journal of Petroleum Science and Engineering 177 (2019) 496–507
2016).
According to Neri (2009), the hydrodynamic dispersion is macro-
scopic outcome of the combined effects of molecular diffusion and
mechanical dispersion, and the transverse dispersion coefficient (DT )
can be generally expressed by the linear combination of the molecular
diffusion term and the mechanical dispersion term:
DT = De + αT |v| (1)
2
where De is the effective diffusion coefficient (m /s), αT is the transverse
dispersivity (m) and v is the interstitial velocity (m/s). De and v are
given by the following equations:
D
De =
τ (2)
u
v=
ϕ (3)
where D is the molecular diffusion coefficient (m /s), τ is the diffusional
2
tortuosity (m2/m2), u is the Darcy velocity (m/s) and ϕ is the porosity.
Measurement of effective diffusion coefficients and dispersivity is
significantly important in that one cannot quantify the solvent effect
properly without accurately measured those values. It is worth men-
tioning that diffusion and dispersion are representing similar concepts
on mass transfer but the former is the physical property describing the
interaction between solute and solvent, and the latter is related to
geological (local) heterogeneity. In this research, we have conducted a
series of experiments using a pressurized bulk fluid container to mea-
sure the diffusion coefficients. Note that the tortuosity of oil sands
should be quantified separately to determine the effective diffusion
coefficients. The measurement of tortuosity and dispersivity is beyond
the scope of this paper but it will be planned in the next step.
Diffusion coefficients between solvents and bitumen have been
discussed by several researchers. According to Ghanavati et al. (2014)
who reviewed the existing researches, there are two major approaches
to define diffusion coefficients. One is concentration independent dif-
fusion coefficients which have been measured by spinning disk, mod-
ified transpiration, Taylor dispersion and nuclear magnetic resonance
(NMR). The other is concentration dependent diffusion coefficients
which have been measured by the pressure decay method, laser or
visible light transmission imaging and X-ray computed tomography
(CT) scanning. The concentration dependent diffusion coefficients are
appropriate to represent the SA-SAGD process because the concentra-
tion dependence becomes significant in liquid-liquid mixing which ac-
tually occurs along the vapor chamber edges in the SA-SAGD and
VAPEX processes. Ghanavati et al. (2014) reports that the order of
diffusion coefficients between Athabasca bitumen and typical liquid
solvents including pentane and hexane at the atmospheric condition is
ranging from 10−11 to 10−10 m2/s. However, to the best of our knowl-
edge, there is no publication about the experiment that can successfully
determine concentration dependent diffusion coefficients between li-
quid solvents and bitumen at high-temperature vapor edge conditions.
In our research, high pressure, high temperature experiments have
been designed and implemented to measure concentration dependent
diffusion coefficients whose order was 10−9 m2/s. Three diffusion ex-
periments were conducted so far. The first and the third experiments
are high temperature (134∘C) and low temperature (76∘C) hexane-bi-
tumen cases, and the second experiment is high temperature (135∘C)
diluent-bitumen case. In this context, diluent is the mixture of various
hydrocarbons that dilute produced bitumen at the surface. The ex-
periments were conducted using a dual-energy X-ray CT scanner.
Generally, it is challenging for medical CT scanners to measure high
temperature systems because CT scanners cannot be housed in a heat
bath. In our research, though, the high pressure, high temperature
diffusion cell which can be four-dimensionally captured by the CT
scanner was manufactured by combining pumps, pipes, valves, line-
heaters and pressurized fluid containers. In fact, it is not very difficult to
M. Imai, et al. Journal of Petroleum Science and Engineering 177 (2019) 496–507

maintain temperature of the system if thermostats and heaters are

properly used, but it is much harder to get an isobaric condition.
The system that we created was similar to what Song et al. (2010)
designed in their research, but our case is high pressure (3.5 MPa) li-
quid-liquid diffusion unlike previously published papers where gas-li-
quid diffusion has been measured at the atmospheric condition. The
biggest challenge of our case is the volume change (typically volume
reduction) due to the liquid-liquid mixing which happens while the
diffusion process is active. Because of the low compressibility of liquids,
tiny volume reduction due to the liquid-liquid mixing causes significant
pressure drop if the experimental system is isolated. This kind of abrupt
and large pressure falling should not be allowed because our assump-
tion is strictly isobaric. Our solution is making a solvent inlet line al-
ways open and controlled so as to maintain pressure using an extra
pump. Thus, our system is now subject to constant pressure-tempera-
ture (PT) condition but technically is not isolated; the diffusion process
can be jammed by fresh solvent influx from the outside of the system.
Since the cumulative influx was still little (a few volume percent total
influx of solvent), measured diffusion profiles looked undisturbed but Fig. 2. Athabasca bitumen used in this experiment (room condition).
attention should be paid on solvent influx which is always active during
the experiment.
and the condensate blend (diluent). The latter one is a mixture of light
hydrocarbons used for diluting bitumen to meet the specification of
2. Estimation of diffusion coefficients
pipeline transportation. At 15∘C, its measured density is 700 (kg/m3). A
picture of the Athabasca bitumen sample used in this research is shown
2.1. Preparation of experiments
in Fig. 2.
2.1.1. Background
Diffusion coefficients between solvents and bitumen have been 2.1.3. Diffusion cell
measured in the context of VAPEX modeling. There are three major Our diffusion experiments happen in a so-called diffusion cell shown
approaches to measure concentration dependent diffusion coefficients. in Fig. 3. The schematic of the diffusion cell is illustrated in Fig. 4.
The first one is the pressure decay method (Etminan et al., 2014). Li- Molecular diffusion between solvents and bitumen progresses in a fluid
quid bitumen contacts gaseous solvents in a transparent container. If chamber. The fluid chamber which is made of a hallow aluminum tube
the experimental environment is pressurized, PVT cells are often used. is confined by end plugs and surrounded by a rubber sleeve which is
Initially, solvent and bitumen are in a closed container. As the mixing pressurized by confining water. The inner diameter of the fluid chamber
induced by inter-phase molecular diffusion advances, the pressure is is 90 mm and the length is 53 mm. The total volume is 337 cm3.
decayed and the interface height between bitumen and solvent rises due The end plugs have multiple ports for different purposes. The left-
to the dissolution of solvent into bitumen. The diffusion coefficients can most end plug has two ports which are the temperature gauge port and
be determined by analyzing time-lapse monitoring data of pressure and the bitumen inlet port. The right-most end plug has only one port which
interface height. This method is convenient in that no special equip- is the solvent inlet/outlet port. The same setting was used for all three
ment is needed except for PVT apparatus. However, the analysis is experiments.
cumbersome because the numerical model which can reproduce one-
dimensional inter-phase molecular diffusion is required. Furthermore,
2.1.4. Experimental setup
this method is only valid for two-phase systems. Since our objective is
The experimental setup and its schematic are shown in Figs. 5 and 6.
to measure diffusion coefficients between two miscible liquids, the
The diffusion cell is mounted on the bed of a medical CT scanner
pressure decay method cannot be used.
(SIEMENS SOMATOM Definition Flash). Our experiments were
The second one is laser or visible light transmission imaging (Oballa
and Butler, 1989). This method measures intensity of light transmission
through the mixture of solvents and bitumen. Diffusion coefficients are
determined such that the time-lapse concentration profiles which are
measured by the light transmission imaging technique are reproduced.
This method can be conducted in a normal PVT equipment with a laser
transmitter and a receiver. However, laser beam cannot go through the
bitumen system. Due to this limitation, laser transmission imaging does
not have sufficient accuracy for the solvent-bitumen systems.
The last one is X-ray CT imaging (Salama and Kantzas, 2005; Luo
et al., 2007). The idea is almost same as the laser transmission imaging
but in this method, X-ray is used instead of laser or visible light. The X-
ray CT imaging is particularly suited for the system consisting of sol-
vents and bitumen because three-dimensional density distribution in a
diffusion cell is clearly captured. We also used the X-ray CT imaging to
capture the diffusive behavior of the solvent-bitumen systems.

2.1.2. Experimental fluids

In our experiments, we used an Athabasca bitumen whose density at
15∘C is 1014.1 (kg/m3). The kinematic viscosity at 50∘C is 9137.3
(mm2/s). For solvents, we used two different fluids: pure normal hexane Fig. 3. Exterior view of diffusion cell.

498
M. Imai, et al. Journal of Petroleum Science and Engineering 177 (2019) 496–507

Table 1
Benchmark case specification.
Sample Temperature °C Pressure MPa, absolute Density kg/m3 CT value

Water 25.0 0.101 997.05 −3.74

Polymer 23.6 0.101 885.48 −107.02
Crude oil 25.0 0.101 768.00 −234.40
n-Decane 25.0 0.101 726.53 −262.33
n-Hexane 25.0 0.101 654.78 −331.03
n-Hexane 137.7 3.51 550.18 −438.77

calculation of diffusion coefficients was negligible because the in-

Fig. 4. Schematic of diffusion cell (not to scale).
filtrated solvent did not reach the region of interest. This issue is dis-
cussed later.
Temperature is maintained by flexible ribbon heaters which are
controlled by thermostats. All of the accumulators, diffusion cell and
connection lines are heated and covered by heat insulation material.
The experimental laboratory is conditioned by a normal air-condi-
tioner. Temperature and pressure data are collected and stored in a
logger PC accordingly for the purpose of the validation of the system
environment.

2.1.5. Correlation between CT values and density

Fig. 5. Actual experimental setup.
conducted in isobaric and isothermal conditions. Pressure is controlled
by multiple pumps. Injection Pump-1 is only used when the bitumen is
transferred from the bitumen accumulator to the diffusion cell. A valve
between diffusion cell and the bitumen accumulator is shut right after
the injection of bitumen is finished. Injection Pump-2 working
throughout the experiments actually controls the system pressure. The
pump is automatically operated so as to maintain a pressure of 3.5 MPa.
The make of pumps is Teledyne ISCO. Two MODEL 260Ds and a
MODEL 500D were used.
Generally, liquid-filled closed vessels may experience huge pressure
change when the liquid volume is slightly changed. To prevent the
sharp pressure drop due to the volume reduction of liquid mixture, all
valves between the diffusion cell and Injection Pump-2 stay open during
the experiments. Pressure is maintained by solvent influx from the right
hand side of the system. The influence of solvent influx that disturbs
What X-ray CT scanners can provide is three-dimensional image of
CT values. Since CT value is a linear function of density, a correlation
between CT value and density can be created. Six samples whose
densities are known were injected into the diffusion cell and were
scanned by the CT scanner to identify their CT values. Table 1 specifies
the conditions to create a correlation. Samples 1 to 5 are measured at an
atmospheric condition, whereas Sample 6 is measured at the condition
of the vapor chamber edge. A standard liquid for calibrating viscometer
(CANNON S8000) was used as a polymer to populate the gap between
pure water and crude oil. As shown in Fig. 7, the measured CT values
and the given density are well correlated.
ρ = 1.0278x ct + 999.59 (4)
where x ct is the CT value, ρ is the density in kg/m3. The coefficient of
determination (R2) was 0.9989. The resulted correlation is used to
convert a CT value image to a density image. Note that the CT value of
Athabasca bitumen is also plotted on Fig. 7, but it is not included for the
creation of the correlation because its density is estimated by a corre-
lation proposed by Gao et al. (2017). In their paper, the density of bi-
tumen is correlated by the following Tait equation which can take into
account pressure and temperature.

Fig. 6. Schematic of experimental setup.

499
M. Imai, et al. Journal of Petroleum Science and Engineering 177 (2019) 496–507

Fig. 7. Correlation between CT value and density.

ρ0 (T , p0 ) Fig. 8. Region of interest and 1D z-axis over the initial CT image of the hexane
ρ (T , p) =
(134∘C) experiment.
1 − β ln ( b+p
b + 0.1 ) (5)

where diffusion cell. A three-dimensional (3D) CT image is generated by

ρ0 = 1149.3967 − 0.3822T − 3.5378 × 10−4T 2 stacking two-dimensional (2D) CT images whose dimensions are 512 ×
(6)
512. The voxel size is 0.218 mm. The slice thickness is 0.6 mm. We have
β = −16.7695 + 0.0578T (7) taken 34 slices for each CT scanning. Fig. 9 is the raw CT image of the
16th slice at the initial condition of the hexane (134∘C) experiment.
b = 3.7568 × 10 4 − 1.8009 × 107T −1 + 2.0605 × 109T −2 (8) A two-dimensional rectangular area is defined as a region of interest
(ROI) which is surrounded by a white rectangular frame in Fig. 8. The
According to Gao et al. (2017), ρ is the density of bitumen in kg/m3,
CT values on the ROI are averaged using slices from No. 6 to 19. On the
ρ0 is the density at atmospheric pressure in kg/m3, T is the temperature
ROI, the one-dimensional (1D) z-axis is defined downward. The ROI has
in K, and p is the pressure in MPa.
81.7 mm height.
Raw data of CT scanning should be processed for the evaluation of
2.2. CT scanning and image processing diffusion coefficients. The objective of image analysis is to obtain the
average CT value profile on the region of interest. Specifically, we
2.2.1. Experiments adopted the following steps to achieve the objective.
Three consecutive diffusion experiments have been conducted. The
first and third experiments measured diffusion coefficients between 1. Pre-processing
normal hexane and Athabasca bitumen at different temperatures (134°C
• To apply the median filter to reduce image noise
and 75°C). The second experiment measured diffusion coefficients be-
tween actual condensate blend (diluent) and Athabasca bitumen at the
• To check the boundary effects by creating contour maps and
profile plots
estimated chamber edge temperature (135°C) of an SA-SAGD process.
Pressure is maintained to approximately 3.5 MPa during all of the ex-
• To conduct segmentation of the images and calculate volume of
the container and fluids
periments. 2. Conversion of the CT value to the density using the correlation
The diffusion cell is initially saturated with solvent. Then bitumen is 3. Extraction of the region of interest (ROI)
injected. Injection duration and injected volume were summarized in
• To determine and extract the ROI
Table 2 for each experiment. CT scanning is consecutively conducted
right after the completion of bitumen injection into the diffusion cell.
• To calculate an average 1D CT value profile of the ROI along the z-
axis
Since the speed of molecular diffusion is decelerated with the square
root of time, the frequency of the CT scanning is gradually reduced.
Note that some mixing is forced during the initial injection process
which may result in the measurement error. To eliminate the error, the
initial time was shifted in the calculation of the diffusion coefficients.

2.2.2. Image analysis workflow

The CT scanner can visualize the distribution of CT values in the

Table 2
Summary of initial fluid injection.
Experiment Solvent Injection Injected bitumen Injected solvent
duration volume volume

sec cm3 cm3

1 Hexane 175 115.7 225.5

2 Diluent 180 114.6 226.2
Fig. 9. Raw CT image of the 16th slice at the initial condition of the hexane
3 Hexane 180 117.6 223.1
(134∘C) experiment.

500
M. Imai, et al. Journal of Petroleum Science and Engineering 177 (2019) 496–507

Fig. 10. Filtering results on CT images of the high temperature hexane experiment (side view: solvent port on the right).

2.2.3. Pre-processing
To smooth the noisy response of the raw CT value profile, the
median filter is applied. A 2D filter was applied because slice thickness
is about 7 times larger than the lateral resolution. In our cases, the
selected radius and iteration number are three voxels and three times,
respectively. The comparison between original and filtered images is
shown in Fig. 10. The filtering effects that can eliminate noises are
clearly observed.

2.3. Calculation of diffusion coefficients

2.3.1. Calculation of concentration profile

According to Fick's first law, the diffusive flux is driven by the molar
concentration gradient. Thus the diffusion coefficient of the Fickian
diffusive flux can be evaluated by analyzing time-lapse profiles of molar
concentration. Fortunately, concentration is somehow correlated with
density, and density profiles are obtained by processing our experi-
mental data.
In general, concentration can be calculated by the following equa-
tion:
ρ − M2/ V2
C1 = m
M1 − M2 V1 / V2 (9)
3
where C is the volumetric molar concentration (mol/m ), ρm is the
mixture density (kg/m3), M is the molecular weight (kg/mol), V̄ is the
partial molar volume (m3/mol) and subscripts 1 and 2 stand for solvent
and bitumen, respectively. If the partial molar volume is constant (i.e.
there is no volume change in mixing of fluids 1 and 2), a simpler ex-
pression shown next can be used.
ρ1 ⎛ ρm − ρ2 ⎞
C1 = ⎜ ⎟
M1 ⎝ ρ1 − ρ2 ⎠ (10)
where ρ1 and ρ2 are the densities of pure solvent and bitumen (kg/m3),
respectively. The derivation of Equations (9) and (10) is summarized in
Appendices A and B.
According to the preliminary mixing experiments where hexane and
bitumen were mixed by 30:70 in weight percent at 3.4 MPa, the actual
mixing between hexane and bitumen does not perfectly behave as an
ideal mixture (Fig. 11). The mean absolute percentage error (MAPE)
between actual mixing and linear ideal mixing was 2.43% which may
lead to the underestimation of the mixture density. Given that error, the
concentration profile used in the following Boltzmann-Matano analysis
introduced in Appendix C may be slightly skewed. Consequently, the
estimated diffusion coefficients may be shifted by a few percents. We
think the error is still trivial, and thus the constant partial molar volume
condition is assumed in this research. Hence, Equation (10) was used.
Fig. 11. Hexane-bitumen mixing experiments.
2.3.2. Assumptions
We made two assumptions to determine diffusion coefficient model.
The first assumption is that mass transport happening near vapor
chamber is approximated by binary diffusion. Binary diffusive flux is
defined between a solvent pseudo component and a bitumen pseudo
component.
The second assumption is that diffusion coefficients have con-
centration dependence as Taylor and Krishna (1993) summarized. For
binary liquid mixtures, concentration dependent diffusion coefficients
are analytically given by the Boltzmann-Matano method (Matano,
1933) which is briefly summarized in Appendix C. In our research, the
Den Broeder approach (den Broeder, 1969; Paul et al., 2014) was used
to determine the diffusion coefficients.
3. Results
3.1. Hexane (high temperature) diffusion experiment
The first experiment was conducted at a temperature of 134°C. The
concentration dependent diffusion coefficients is shown in Fig. 12. Note
that normalized solvent concentration (CD ) is calculated by:
CD = C1/ C1, pure (11)
3
where C1 is the solvent molar concentration in mixture (mol/m ) and
C1, pure is the pure solvent molar concentration (mol/m3).

501
M. Imai, et al. Journal of Petroleum Science and Engineering 177 (2019) 496–507

Fig. 12. Diffusion coefficients (hexane, 134°C). Fig. 14. Diffusion coefficients (hexane, 75°C).

The averaged diffusion coefficient which is calculated by averaging

data obtained from 3 h to 9 h is illustrated as a solid black line. The
profile of concentration dependent diffusion coefficients is supposed to
be identical regardless of time. However, in the first several time steps,
the diffusion coefficients are deviated from the averaged curve. This
deviation is caused by the disturbance during the injection process
which evokes convective mixing.

3.2. Diluent (high temperature) diffusion experiment

The second experiment was conducted at a temperature of 135°C

using a condensate blend (diluent) as solvent. The profile of diffusion
coefficients is shown in Fig. 13. The averaged diffusion coefficient be-
comes slightly smaller than that of the hexane case. At the last time step
(7200 min), phase separation was observed. The density of the sepa-
rated phase is about 921 kg/m3 that is slightly lighter than the initial
bitumen density (940 kg/m3). Fig. 15. Diffusion coefficients (all experiments).

3.3. Hexane (low temperature) diffusion experiment solvent type and temperature. The overall tendency is that:

The third experiment was conducted at a temperature of 75°C to

examine temperature dependence of diffusion coefficients. The profile
• all of the experiments show dependence of diffusion coefficients on
concentration in a similar manner;
of diffusion coefficients is shown in Fig. 14. As expected, the averaged
diffusion coefficient becomes smaller than that of the high temperature
• hexane and diluent have almost same diffusion coefficients at the
vapor chamber condition (135°C, 3.5 MPa), but diluent shows
case. slightly smaller ones;

3.4. Dependence of diffusion coefficients

• a clear dependence on temperature was observed in the hexane
experiments;

Fig. 15 shows dependence of diffusion coefficients on concentration,

• diffusion coefficients become about a half, when temperature is
decreased from 407 K (134°C) to 348 K (75°C).

3.5. Phase separation

In the first and second experiments, we experienced unexpected

Fig. 13. Diffusion coefficients (diluent, 135°C).
phase separation in the diffusion cell. A previously reported fluid
characterization research using hexane and the Athabasca bitumen
(Gao et al., 2018) does not observe this kind of phase separation. In
other words, hexane and bitumen are supposed to be miscible. How-
ever, in the high temperature hexane experiment, small amount of the
second phase was observed on the right hand side of the diffusion cell
after 118 h (Fig. 16). Similarly, in the diluent case, the second phase
emerges after 120 h (Fig. 17). In the low temperature hexane case,
phase separation was not detected during the experiment, but small
amount of separated phase was observed along the right most wall (the
end plug) when the system temperature was reduced to 26°C (Fig. 18).
The asphaltene-like solid state was actually observed when the diffu-
sion cell was dismantled. Since the point of deposition is particularly
close to the inlet point of fresh solvents, it seems that asphaltene was

502
M. Imai, et al. Journal of Petroleum Science and Engineering 177 (2019) 496–507

Table 3
Summary of experimental conditions and averaged diffusion coefficients.
Solvent Pressure MPa, Temperature °C Diffusion coefficient (D) m2/
absolute s

Hexane 3.51 134.2 1.47× 10−9

Diluent 3.51 135.3 1.23× 10−9
Hexane 3.44 75.3 0.72× 10−9

Table 4
Summary of phase separation.
Solvent Elapsed Separated Separated Surrounding

time phase volume phase density phase density

Fig. 16. Phase separation (high temperature hexane case). 3 3
hours cm kg/m kg/m3

Hexane (134°C) 118 0.27 912 783

Diluent (135°C) 120 8.33 921 821
Hexane (75°C) 281 (0.34) (857) (757)

experiment, whereas our hexane diffusion coefficient at 75∘C increases

by a factor of two. Considering the temperature dependence observed
in a series of experiments, our diffusion coefficients seems to be within
a reasonable range.
Table 4 summarizes the observed phase separation. Note that values
for the low temperature (75°C) hexane case are in brackets because
density and volume of separated phase may not be accurate due to
partial volume effect in CT scanning. The largest phase separation was
observed in the diluent case.

Fig. 17. Phase separation (high temperature diluent case).

4. Discussion

4.1. Concentration dependence on diffusion coefficients

Our experimental results clearly exhibit the concentration depen-

Fig. 18. Phase separation (low temperature hexane case).
precipitated because of the contact of existing bitumen and substantial
amount of fresh solvents. During the cooling period, about 40 cm3 of
hexane was supplied to the system in the low temperature hexane ex-
periment.
3.6. Summary of major findings
The actual experimental conditions and the corresponding averaged
diffusion coefficients are summarized in Table 3. As already mentioned,
diffusion coefficients have dependence on solvent type and temperature
as well as concentration. The diffusion coefficients measured in our
experiments are reasonably larger than the diffusion coefficients mea-
sured by Salama and Kantzas (2005). For instance, the average diffu-
sion coefficient between hexane and bitumen at room temperature was
3.52× 10−10 m2/s (average over the one-week experiment) in their
dence on mutual diffusion coefficients. As shown in Fig. 15, the diffu-
sion coefficient profiles demonstrate the nonlinear bimodal trends. The
first peak appears in the vicinity of the pure solvent region and the
second peak appears in the mixing zone where the solvent concentra-
tion is about 0.4–0.5. Salama and Kantzas (2005) showed the similar
bimodal profiles in their measurements of mutual diffusion coefficients
between hexane and bitumen at low temperature conditions using the
X-ray CT scan technique.
Nonlinear concentration dependent profiles have been reported in
earlier literature. A theoretical explanation was given for example in
Taylor and Krishna (1993) where strong concentration dependence of
diffusion coefficients was justified by specifying the thermodynamic
factor Γ :
Γ= Γ(x1, x2) (12)
where x1 and x2 are the mole fractions of components 1 and 2 in binary
systems. Since the Fickian mutual diffusion coefficient D̃ is the product
of Γ and the Maxwell-Stefan coefficient, the resulted Fick D̃ shows
strong concentration dependence.
In the meanwhile, Oballa and Butler (1989) suggested that the bi-
modal profile might be related to asphaltene precipitation. They pre-
sumed that the second peak on the profile was brought by the enhanced
solvent (toluene) mobility due to the transient asphaltene deposition
though eventually toluene and bitumen formed the miscible state. In
our experiments, no explicit relationship between separation of the
extra phase and nonlinearity of the bimodal profiles was observed.
However, given the fact that asphaltene precipitation was implied in all
experiments, we cannot exclude the possibility that the bimodal profile
represents the effect of the asphaltene precipitation.

503
M. Imai, et al.
Fig. 19. Hexane supply volume of low temperature hexane experiment.
4.2. Effect of solvent influx during experiments
Our diffusion experiments are not strictly isolated. To prevent the
abrupt pressure drop due to the shrinkage of fluid by liquid-liquid
mixing, the solvent inlet port was kept open during the experiments.
Consequently, certain amount of solvents (hexane and diluent) was
supplied to the system. The volume of supplied solvent in the low
temperature hexane experiment is 1.9% (Fig. 19). The curve on the
figure is steadily increased as the mixing (diffusion) advanced, though
the curve was slightly oscillated due to the fluctuation of the room
temperature. Under such circumstances, supplied solvent may jam the
spontaneous diffusion behavior, in the worst case scenario.
However, our ROIs looked unsusceptible by the solvent influx (see
Fig. 10). Since the supply point of solvent is located at the center of the
rightmost end plug, the concentration counter must be jammed and an
anomaly could be seen near the supply point if the effect of supplied
solvent is dominant. However, we only observed the monotonous
concentration counter maps which can be seen in the case of the ideal
diffusion. We think that our ROI is far enough from the supply point in
view of the speed of solvent transport. At this point, we concluded that
the effect of solvent influx during diffusion experiments could be ne-
glected.
4.3. Effect of inertia force at each CT scanning opportunity
Our experimental system was mounted on the bed of the medical CT
scanner. The system slightly accelerates and decelerates when a CT
image is taken because the bed physically moves for a shooting. In that
case, the experimental system is subject to the inertia force which
perturbs the system. The mixing speed may be accelerated due to this
perturbation.
Appendix D. Supplementary data
Supplementary data to this article can be found online at https://doi.org/10.1016/j.petrol.2019.02.049.
Appendix A. Derivation of molar concentration of binary system in general expression
According to Taylor and Krishna (1993), the concentration measures that describe binary mixtures are generally given by the following equa-
tions:
V¯m = x1 V¯1 + x2 V¯2
1
V¯m =
Cm
Ci
xi =
Cm
504
Journal of Petroleum Science and Engineering 177 (2019) 496–507
Unfortunately, we do not have an effective way to identify the effect
of the inertia force that may or may not affect the diffusion behavior so
far. However, at least, our diffusion coefficients are in a good range
with respect to the reference values which are, for example, reported by
Ghanavati et al. (2014). Twhe inertia force may not affect the diffusion
process significantly because of the active viscous force working on our
experimental fluids which move steadily with the system such that the
effect of inertia is canceled.
We suppose considering inertia is not necessarily needed, but to
identify the effect of inertia, we can conduct another experiment which
is identical to the original one except for the frequency of CT scanning.
An experiment with less frequent CT scanning opportunities must have
a meaningfully slower mixing speed in the case where the effect of
inertia force is considerable.
5. Conclusion
We conducted three consecutive diffusion experiments in which
concentration dependent diffusion coefficients between solvents (pure
hexane or diluent) and bitumen were identified. We fixed a pressure
around 3.5 MPa for each experiment. Hexane was used as a solvent in
the first and the third experiments. The experimental temperatures
were 134.2∘C for the first one, 135.3∘C for the second one and 75.3∘C for
the third one. In the second experiment, diluent was used instead of
pure hexane so as to see the effect of solvent species.
The order of averaged diffusion coefficient was reasonable, and the
values range from 0.72× 10−9 m2/sec (hexane, 75.3∘C) to
1.47× 10−9 m2/sec (hexane, 134.2∘C). A slight sensitivity of solvent
species is observed, but not significant (1.23× 10−9 m2 for the diluent-
bitumen system at a temperature of 135.3∘C). Hexane has slightly
higher diffusion coefficients than diluent. There is clear sensitivity in
temperature. The comparison of the first and third experiments in-
dicates that the same solvent has higher diffusion coefficients in a
higher temperature condition.
Through the experiments that we conducted in this research, we
successfully characterized the diffusion coefficients in the SA-SAGD
process. For the next step, the evaluation of tortuosity and dispersivity
is expected.
Acknowledgements
We thank Japan Oil, Gas and Metals National Corporation
(JOGMEC), Japan Petroleum Exploration Co., Ltd. (JAPEX) and Japan
Canada Oil Sands Limited (JACOS) for the approval of publishing this
paper. We also appreciate the technical supports from Mr. Tanetomo
(Tom) Izumi, Mr. Yutaro Kaito (JAPEX), Mr. Tatsuya Hattori, Ms.
Chiaki Otomo, Dr. Makoto Kunieda and Dr. Hirofumi Okano
(JOGMEC).
(A.1)
(A.2)
(A.3)
M. Imai, et al. Journal of Petroleum Science and Engineering 177 (2019) 496–507

ρi
Ci =
Mi (A.4)
ρm = ρm,1 + ρm,2 (A.5)
3 3
where V̄m is the mixture molar volume (m /mol), V̄i is the partial molar volume of i (m /mol), x i is the mole fraction of i, Cm is the mixture molar
concentration (mol/m3), Ci is the molar concentration of i (mol/m3), Mi is the molecular weight of i (kg/mol), ρm is the mixture mass density (kg/m3)
and ρm, i is the mass density of i in the mixture (kg/m3).
Using Equations (A.1) to (A.3), we can lead to the following expression:
C1 V¯1 + C2 V¯2 = 1 (A.6)
Plugging Equation (A.4) in Equation (A.6), we can eliminate C2 :
ρ2
C1 V¯1 + V¯2 = 1.
M2 (A.7)
Then ρm,2 can be eliminated using Equation (A.5):
V¯2
C1 V¯1 + (ρm − ρ1) = 1.
M2 (A.8)
Again, using Equation (A.4), we eliminate ρm,1:
V¯
C1 V¯1 + (ρm − C1 M1) 2 = 1.
M2 (A.9)
By solving Equation (A.9) for C1, the final expression is obtained.
ρm − M2/ V2
C1 =
M1 − M2 V1 / V2 (A.10)
This is the general expression of concentration for binary mixtures.

Appendix B. Derivation of molar concentration of binary system without volume change in mixing

Equation (A.10) is always true in the binary mixing. However, the following equations are satisfied when there is no volume change in mixing of
fluids 1 and 2:
M
V¯1 = 1
ρ1 (B.1)
M
V¯2 = 2
ρ2 (B.2)
3
where ρi is the mass density of pure i (kg/m ).
Using these relations, we can eliminate the partial molar volumes V̄1 and V̄2 from Equation (A.10):

ρ1 ⎛ ρm − ρ2 ⎞
C1 = ⎜ ⎟.
M1 ⎝ ρ1 − ρ2 ⎠ (B.3)

Appendix C. Boltzmann-Matano method

In the assumption of concentration dependent diffusion coefficients, the one-dimensional binary diffusion equation is given by:
∂Ci ∂ ⎛ ∂Ci ⎞ ∂
= Di − (vCi ) i = 1,2
∂t ∂z ⎝ ∂z ⎠ ∂z (C.1)
where Di is the intrinsic diffusion coefficient and v is the flow velocity along the z-axis. The first term on the right hand side is the diffusion term
(Fick's first law) and the second term is the superficial convection term which is brought by the difference between D1 and D2 and the volume change
due to mixing. If the partial molar volume is constant (i.e. the system volume is unchanged due to mixing), the above equation can be written using
mutual diffusion coefficient D̃ and solvent concentration C = C1.
∂C ∂ ⎛ ˜ ∂C ⎞
= D
∂t ∂z ⎝ ∂z ⎠ (C.2)
where
D˜ = C2 V2 D1 + C1 V1 D2 (C.3)
The partial differential equation (PDE) shown in Equation (C.2) can be converted to the ordinary differential equation (ODE) by Boltzmann's
transformation with introducing a new parameter B.
z
B=
t (C.4)
Using B, the given PDE can be transformed as follows:

505
M. Imai, et al. Journal of Petroleum Science and Engineering 177 (2019) 496–507

B dC d ⎛ ˜ dC ⎞
− = D
2 dB dB ⎝ dB ⎠ (C.5)
The above expression is further simplified to the total derivative form because dB can be canceled:
B dC ⎞
− dC = d ⎛D˜
2 ⎝ dB ⎠ (C.6)
In our experimental condition, the initial and boundary conditions (IC and BCs) are given by the simple expressions if the initial contact position
between solvent and bitumen is zero:

• IC:
C (z , 0) = 1 z < 0 (C.7)

C (z , 0) = 0 z > 0 (C.8)

• BCs:
C (B = −∞) = 1 (C.9)

C (B = ∞) = 0 (C.10)
At an arbitrary concentration C, we can find that the following equation is satisfied by integrating Equation (C.6) from C′ = 0 to C′ = C .
C −B dC
∫0 2
dC′ = D˜
dB
|C
(C.11)
Obviously,
dC
=0
dB (C.12)
is satisfied at C = 0 .
If the original variables are used instead of B, the final form becomes:
−1 C dC
2 t
∫0 zdC ′ = t D˜
dz (C.13)
C

which results in:

1 dz˜ C
D˜ = −
2t dc
∫0 ˜ ′
zdC
(C.14)
C

where, z̃ is the shifted coordinate that is associated with the Matano interface zM .
1 1
∫0 ˜
zdC = ∫0 (z − zM ) dC = 0 (C.15)
The Matano interface indicates the equal-area point can be interpreted as shown in Fig. C1. In our research, the Matano interface was captured by
our in-house Matlab program.

Fig. C.1. Matano interface.

References Steam/Solvent Processes for the Recovery of Heavy Oil and Bitumen. Society of
Petroleum Engineers.
Bayestehparvin, B., Ali, F.M.S., Abedi, J., Aug. 2018. Solvent-Based and Solvent-Assisted
Ardali, M., Barrufet, M.A., Mamora, D.D., Qiu, F., Jan. 2012. A Critical Review of Hybrid Recovery Processes: State of the Art. SPE Reservoir Evaluation & Engineering.

506
M. Imai, et al. Journal of Petroleum Science and Engineering 177 (2019) 496–507

Butler, R.M., Feb. 1998. Thermal Recovery of Oil and Bitumen. GravDrain Inc., Calgary, Salama, D., Kantzas, A., Jan. 2005. Monitoring of Diffusion of Heavy Oils with
Alberta. Hydrocarbon Solvents in the Presence of Sand. Society of Petroleum Engineers.
Butler, R.M., Mokrys, I.J., 1989. Solvent analog model of steam-assisted gravity drainage. Song, L., Kantzas, A., Bryan, J.L., Jan. 2010. Experimental Measurement of Diffusion
AOSTRA J. Res. 5 (1), 17–32. Coefficient of CO2 in Heavy Oil Using X-Ray Computed-Assisted Tomography under
Chen, Z.J., Keshavarz, M., 2015. Meeting the Challenges in the Post-steam Era. Reservoir Conditions. Society of Petroleum Engineers.
den Broeder, F., May 1969. A general simplification and improvement of the matano- Taylor, R., Krishna, R., 1993. Multicomponent Mass Transfer. Wiley, New York; Toronto,
Boltzmann method in the determination of the interdiffusion coefficients in binary oCLC, pp. 883370360.
systems. Scripta Metall. 3 (5), 321–325. Zhao, L., Jan. 2004. Steam Alternating Solvent Process. Society of Petroleum Engineers.
Etminan, S.S.R., Maini, B.B., Chen, Z.J., Jun. 2014. Modeling the Diffusion Controlled
Swelling and Determination of Molecular Diffusion Coefficient in Propane-Bitumen
System Using a Front Tracking Moving Boundary Technique. Society of Petroleum Nomenclature
Engineers.
Gao, J., Okuno, R., Li, H.A., Jun. 2017. An experimental study of multiphase behavior for B: Parameter for Boltzmann transformation
n-butane/bitumen/water mixtures. SPE J. 22 (03), 783–798. b: Parameter for Tait equation
Gao, J., Okuno, R., Li, H.A., Feb. 2018. A phase-behavior study for n-hexane/bitumen and C: Molar concentration [mol/m3 ]
n-octane/bitumen mixtures. SPE J. 23 (01), 128–144. CD : Normalized (dimensionless) concentration
Ghanavati, M., Hassanzadeh, H., Abedi, J., Aug. 2014. Critical review of mutual diffusion D: Diffusion coefficient [m2 /s]
coefficient measurements for liquid solvent+bitumen/heavy oil mixtures. Can. J. DT : Transverse dispersion coefficient [m2/s]
Chem. Eng. 92 (8), 1455–1466. De : Effective diffusion coefficient [m2/s]
Gupta, S., Gittins, S., Picherack, P., Nov. 2005. Field implementation of solvent aided Di : Intrinsic diffusion coefficient of i [m2/s]
process. J. Can. Petrol. Technol. 44 (11). D̃ : Mutual diffusion coefficient [m2/s]
Leaute, R.P., Jan. 2002. Liquid Addition to Steam for Enhancing Recovery (LASER) of M: Molecular weight [kg/mol]
Bitumen with CSS: Evolution of Technology from Research Concept to a Field Pilot at p: Pressure [MPa]
Cold Lake. Society of Petroleum Engineers. p0 : Atmospheric pressure [MPa]
Luo, H., Kryuchkov, S., Kantzas, A., Jan. 2007. The Effect of Volume Changes Due to T: Temperature [K]
Mixing on Diffusion Coefficient Determination in Heavy Oil and Hydrocarbon Solvent t: Time [s]
System. Society of Petroleum Engineers. u: Darcy velocity [m/s]
Matano, C., 1933. On the relation between the diffusion-coefficients and concentrations V̄i : Partial molar volume of i [m3/mol]
of solid Metals (the nickel-copper system). Jpn. J. Phys. 8, 109. V̄m : Mixture molar volume [m3/mol]
Nasr, T.N., Beaulieu, G., Golbeck, H., Heck, G., 2003. Novel expanding solvent-SAGD v: Interstitial velocity [m/s]
process” ES-SAGD”. J. Can. Petrol. Technol. 42 (01). xi : Mole fraction of i
Neri, M.G., 2009. Aspects of Transverse Dispersion in Porous Media. Ph.D. thesis. x ct : CT value
Utrecht University. z: Coordinate in z-direction [m]
Oballa, V., Butler, R.M., 1989. An experimental study of diffusion in the bitumen-toluene zM : Location of Matano interface [m]
system. J. Can. Petrol. Technol. 28 (02). αT : Transverse dispersivity [m]
Ogino, K., 2004. Development of Canada oilsands further challenges. J. Jpn. Assoc. Pet. β: Parameter for Tait equation
Technol. 69 (6), 612–620. ϕ: Porosity
Orr, B., Jan. 2009. ES-SAGD; Past, Present and Future. Society of Petroleum Engineers. Γ : Thermodynamic factor
Paul, A., Laurila, T., Vuorinen, V., Divinski, S.V., Jul. 2014. Thermodynamics, Diffusion ρ: Density [kg/m3 ]
and the Kirkendall Effect in Solids. Springer. ρ0 : Density at atmospheric pressure [kg/m3]
Rabiei Faradonbeh, M., Harding, T.G., Abedi, J., Nov. 2016. Semi-analytical modeling of τ: Tortuosity [m2 /m2]
steam-solvent gravity drainage of heavy oil and bitumen: steady state model with
linear interface. Fuel 183, 568–582.

507