Beruflich Dokumente
Kultur Dokumente
net/publication/329448323
CITATIONS READS
0 89
6 authors, including:
A.P. Sasmito
McGill University
121 PUBLICATIONS 1,246 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by A.P. Sasmito on 10 April 2019.
a
Department of Mining and Materials Engineering Department, McGill University, 3450
University Street, Montreal, QC Canada H3A 0E8
b
Mechanical Engineering Department, Indian Institute of Technology, Kanpur, India
1
Pressure loss and friction factor in non-Newtonian mine paste backfill:
modelling, loop test and mine field data
Abstract
The prediction of pressure loss along paste backfill pipeline system requires
friction factor correlations are available in literature to predict pressure loss for paste;
however, none of them was specifically developed and validated for mine paste backfill.
This study addresses the selection of friction factor correlations in hydraulic model to be
tested and validated against pressure loss from laboratory scale pipe loop measurement,
in-situ mine field data and computational fluid dynamics model. Statistical analysis was
performed to quantify the errors. The results suggest that Swamee and Aggarwal friction
factor correlation can accurately predict the pressure loss along the paste backfill pipeline
Keywords: paste backfill; friction factor; rheology; computational fluid dynamics; non-
Newtonian; Bingham.
1. Introduction
Backfilling is a mine waste management method that sends the tailings back
underground to provide stability to the mine walls for successful extraction from adjacent
stopes [1]. The open stope voids are filled with backfill to provide stability for the
adjacent work areas and reducing the risk of local and regional ground failure from
collapse and caving in of the structure. Backfill is typically made from waste rock or
dewatered tailing residues and is often mixed with cement to achieve moderate strengths.
It is delivered to stopes via trucks or by pumping a dense tailings slurry or paste through
boreholes and pipelines. The backfill is made by mixing tailings (sometimes blended with
sand or other aggregate), water and a binder to produce the desirable structural properties
2
[1][2]. The fine tailings are mixed with small amounts of water to enable transport by
pipeline without producing a settling slurry. The resulting fluid is a thick paste with high
mass concentration (usually greater than 70 weight % solids), with a minimum fines
[3][4][5][6][7]. The amount of water added is controlled to provide a balance between the
backfill fluidity and its mechanical properties which is one of its key performance
properties [8].
Backfill is transported from the processing plant on the surface into the mine via
pipelines under gravitational force or pumping pressure [9]. The pressure variation along
the pipe length is important in the pipeline design when evaluating the need for pumps
and booster stations in a pipeline and in the selection of the optimum pipe size. A key
parameter in the pressure loss is the friction factor which is commonly expressed by the
Fanning friction factor or the Darcy-Weisbach friction factor that differ from each other
by a factor of 4 as shown in Eqs. (12) and (13). Friction factor correlations for dilute and
moderately dense slurry flows for Newtonian and non-Newtonian fluids have been
extensively studied in past literature while studies involving high concentration slurry
flows are relatively recent. It is the friction factor for high concentration, non-Newtonian
Senapati and Mishra simulated 60 to 70% by weight fly ash slurry using the non-
Newtonian power law model and define the friction loss using the fanning friction factor
[10]. Liu et al studied the rheology of 50 to 70% by weight coal sludge slurry [11]. They
used the Herschel-Bulkley model to study the rheology of their flow but make no mention
of friction factor. Assefa and Kaushal did a comparative study of several friction factor
correlations for smooth pipes for 60 to 70% by weight slurry using experimental data
3
the best in agreement with experimental data for the Reynolds number range 200 to
40,000. Wu simulated 70 to 80% by weight coal gangue-fly ash slurry and compared it
with loop test experiments [4][17]. In their study the authors employed Swamee &
Aggarwal’s friction factor model [18]. Despite there being a few studies tackling high
concentration slurry flows, none of them focus on the friction factors at low Reynolds
simulations were found together for Bingham plastic fluids. The only friction factor study
which compared some of these elements was done by Assefa and Kaushal, but the study
doesn’t focus on high concentrations at low Reynolds numbers, which is very typical of
Qi et al. experimentally observed the pressure drop in pipe flow of cemented paste
backfill in complex circuits and developed a prediction model using gradient boosting
regression tree (GBRT) [19]. The authors achieved a 0.999 and 0.998 correlation
coefficient and declared the circuit shape as the most sensitive variable followed by solid
correlations in the low Reynolds numbers regime for non-Newtonian flows. They are in
turn compared with experimental data and numerical results with the objective of finding
the best suited friction factor relationship for this flow regime. The experimental data are
obtained from a laboratory-scale flow loop experiment using tailings and mine process
water from two mines in Canada, henceforth referred to as Mine-A and Mine-B. The
numerical study is performed using CFD by adopting the two-parameter Bingham plastic
number as proposed by K. Madlener, B. Frey and H.K. Ciezki [20] against the Newtonian
4
Reynolds number that is conventionally used for non-Newtonian fluids is done to assess
rheology of the backfill. For the fluid model, the rheological parameters are determined
by curve fitting the loop test pseudo shear results with a non-Newtonian fluid equation.
There are several Non-Newtonian fluid models, but the Herschel-Bulkley model and it’s
subgroup, the Bingham plastic model, are found to best represent the nature of high
concentration slurries such as paste backfill [21]. Though it has been observed that in
some cases the three parameter Herschel-Bulkley model better fits the fluid rheogram
[6][22], the Bingham plastic model is more popular because of its simplicity
[4][5][17][23] and it is used in this study to determine the rheology of paste backfill.
The composition of the tailings from Mine-A and Mine-B are detailed below in
Table 1. The particle size distribution of tailings is shown in Fig.1 and Table 2. Mine-A
tailings has a wider range of particle sizes than Mine-B. Mine B tailings are very fine
with 80% of the material passing 20 µm. Particle sizes have significant impact on the
pressure drop across a pipe and is dependent on solid fraction and flow rate. Finer
particles have low sediment velocity and can be easily kept in circulation in comparison
the length of the pipe and characterize the fluid flow behaviour in pipes. A schematic
arrangement of the setup is shown below in Fig. 2. The system is in total 12 m long with
pipes running 8.2 m and the rest comprising of the positive displacement pump and
5
hopper. The pipe inner diameter is 0.045 m (1.83 in). The temperature of the backfill in
the pipe is controlled via a heat exchanger which is connected to a temperature control
mechanism. The mine tailings and water are homogeneously mixed in the hopper and
introduced into the system. The positive displacement pump circulates the fluid using a
variable frequency drive (VFD) to control the mass flow rate of the fluid.
There are 4 pressure gauges installed; 2 upstream and 2 downstream. The pipe
length L between gauges P1-P2 and P3-P4 is 1500 mm. From the data collected by the
pressure gauges, the pressure drop due to the resistance in flow is obtained by removing
the gravitational head loss as detailed below in Eqs. (1) and (2).
Applying Eqs. (1) and (2) to the loop test pressure readings, it was observed that
ΔP12 was equal to ΔP34 within an error of 0.1%. A single value of pressure gradient is
obtained by averaging ΔP12 and ΔP34 as shown below in Eq. (3). Prior to collecting the
pressure data, the backfill is permitted to circulate for sufficient time till a steady-state
∆𝑃 1 ∆𝑃12 ∆𝑃34
= ( + ) (3)
𝐿 2 𝐿 𝐿
The backfill is first mixed at the thickest desired concentration and brought to the
desired temperature using the heat exchanger. It is pumped at various speeds using the
VFD during which time the pressures in the loop are recorded along with the flowrate.
The paste is diluted by adding processed mine water and the test is repeated for lower
oven to determine the exact solid fraction of the backfill used in each trial. Before and
6
after the completion of each test, water is flushed through the pipes to clean out residues
and settlements from previous trials. This entire process is repeated at three different
For determining the rheological properties of the paste backfill, the Buckingham-
Reiner equation shown below in Eq. (4) is used. The shear rate 𝛾 ′ from this is equated
with the pseudo shear rate in Eq. (5). The wall shear stress expressed in terms of pressure
loss is displayed in Eq. (6). The values from Eqs. (5) and (6) are curve-fit into Eq. (7)
using linear regression to determine the Bingham plastic viscosity 𝜇𝑃 and the Bingham
yield stress 𝜏𝐵 .
𝜏𝑊 4 𝜏𝐵 1 𝜏𝐵 4
𝛾′ = [1 − ( ) + ( ) ] (4)
𝜇𝑃 3 𝜏𝑊 3 𝜏𝑊
8𝑉
𝛾′ = (5)
𝐷
𝐷 ∆𝑃
𝜏𝑊 = (6)
4 𝐿
4. Reynolds Number
The dimensionless Reynolds number is the ratio of inertial forces to viscous forces
experienced by a fluid while flowing. In practice it is used to identify the flow regime as
laminar, transitional or turbulent. If the inertial forces, which resist a change in velocity
is dominant then the flow is turbulent and if the viscous forces, which provide a resistance
to the flow is dominant, then the flow is laminar. In an internal flow through a pipe of
hydraulic diameter 𝐷, the Newtonian Reynolds number is written as shown below in Eq.
(7).
𝜌𝑢𝐷
𝑅𝑒𝑛𝑒𝑤𝑡 = (7)
𝜇
7
However, this relies on the assumption that the dynamic viscosity 𝜇 is constant.
This is not the case in non-Newtonian fluids. In 1955, Metzner and Reed developed the
𝜌𝑢2−𝑛 𝐷𝑛
𝑅𝑒𝑔𝑒𝑛 = (8)
8𝑛−1 𝜇𝑝
For Newtonian fluids, 𝜇𝑝 = 𝜇 and 𝑛=1 and Eq. (8) reduces to Eq. (7). For Bingham
plastic fluids, the Eq. (8) reduces to the Bingham Reynolds number.
𝜌𝑢𝐷
𝑅𝑒𝐵 = (9)
𝜇𝑝
the generalized Reynolds number for Herschel-Bulkley type fluids. The authors
compared the results with paraffin-based fluids and noted good agreement between
experimental data and theoretical Darcy friction factor for 𝑅𝑒𝑔𝑒𝑛 𝐵 =1000. At higher
Reynolds numbers lower theoretical friction factor values were observed. The generalized
𝜌𝑢𝐷
𝑅𝑒𝐺𝐸𝑁 𝐵𝐼𝑁𝐺𝐻𝐴𝑀 = (10)
𝜏 𝐷 (3𝑚 + 1)
( 8𝐵 ) ( 𝑢 ) + 𝜇𝑝
(4𝑚)
8𝑢
𝜇𝑝 ( 𝐷 )
𝑚= (11)
8𝑢
𝜏 𝐵 + 𝜇𝑝 ( 𝐷 )
Like the Reynolds number, the Hedstrom number is a measure of the regime
characterizing a flow. It is a dimensionless value that combines the yield stress to viscous
stress relationship of Bingham number with the flow properties of Reynolds number. At
𝐻𝑒 = 0 the fluid is Newtonian and it increases proportionately with Bingham yield stress.
𝜌𝑢𝐷 𝜏𝐵 𝐷 𝜌𝐷2 𝜏𝐵
𝐻𝑒 = 𝑅𝑒𝐵 × 𝐵 = [ ]×[ ]= (12)
𝜇𝑝 𝜇𝑝 𝑢 𝜇𝑝 2
8
It is typically used to calculate a flow’s critical Reynolds number to represent the
point at which the change from laminar to turbulent flow occurs. In the work by Swamee
and Aggarwal [18], the authors arrive at a relationship between critical Reynolds number
and Hedstrom number for the range, 1 ≤ 𝐻𝑒 ≤ 108 , shown below in Eq. (13).
𝐻𝑒 0.35
𝐻𝑒 = 2100 [1 + ] (13)
3600
5. Friction Factor
The concept of friction factor originated from the phenomenological Darcy-
𝛥𝑃 2𝐷
𝑓= (14)
𝐿 𝜌𝑢2
𝛥𝑃 𝐷
𝑓= (15)
𝐿 2𝜌𝑢2
The first non-Newtonian friction factor study in the laminar regime was
completed by Metzner and Reed for different fluids with their rheology modelled by the
power law [25]. A study performed by Garcia and Steffe [26], compared friction factors
of different kinds of non-Newtonian fluids, including Bingham fluids, and concluded that
the results from friction factor models available at that time differed significantly. As the
study focused on higher Reynolds numbers and lacked experimental backing it is not
directly applicable to the present study. The most recent comparative friction factor study
was carried out by Assefa and Kaushal [12]. They reviewed many of the available friction
factor models for non-Newtonian fluids including Hagen-Poiseuille [27], Blasius [28],
Buckingham-Reiner [29], Colebrook [30], Prandtl [31], Darby and Melson [32], Slatter
9
[33], Danish & Kumar [34], Swamee and Aggarwal [18], Wilson and Thomas [15] and
Morrison [16]. But, the study doesn’t focus on high concentration slurries at low flow
rates that represent paste backfill flow. Consequently, the present study was initiated to
compare the relevance of friction factor models for Bingham fluids at high concentration
and low flow rates, using the Assefa and Kaushal study as a starting point for statistical
The friction factor models discussed in this study are those put forth by Hagen-
Poiseuille [27], Swamee and Aggarwal [18], Danish & Kumar [34], J.S. Curtis [35],
Darby and Melson [32], Ihle and Tamburrino [36] and Hanks [37]. The friction factor
models are discussed in detail below starting with the Buckingham-Reiner equation.
5.1. Buckingham-Reiner
An exact description of Darcy-Weisbach friction loss for Bingham plastics for a
fully developed flow was first published by E. Buckingham [29]. The Eq. (16) below is
written in a dimensionless form using the Bingham Reynolds and Hedstrom numbers.
Solving this equation for an exact analytical solution is complex due to the fourth order
64 1 𝐻𝑒 64 𝐻𝑒 4
𝑓= [1 + − ( 3 7 )] (16)
𝑅𝑒𝐵 6 𝑅𝑒𝐵 𝑅𝑒𝐵 𝑓 𝑅𝑒𝐵
5.2. Hagen-Poissuille
This is the simplest and most common formula used to determine friction factor
The formulation is simple and fits accurately for the laminar flow of Newtonian fluids.
10
The model ignores the Bingham yield stress and relies solely on the Bingham plastic
viscosity.
64
𝑓= (17)
𝑅𝑒𝐵
plastic fluids flowing in pipes in the laminar flow regime. It is an approximation of the
avoiding the use of complex and time-consuming iterative solutions. The model is a good
solution to the principal Buckingham equation and depending on how closely a fluid
shows Bingham behaviour, this model can be used. An exact Bingham fluid would be
exceedingly well governed by this model but the discrepancy in our case, though less
compared to other models, is due to the approximation that the backfill is modelled as a
Bingham fluid.
𝐻𝑒 1.143
64 10.67 + 0.1414 (𝑅𝑒 )
𝐵
𝑓= + 1.16
(18)
𝑅𝑒𝐵 𝐻𝑒
(1 + 0.0149 (𝑅𝑒 ) ) 𝑅𝑒𝐵
𝐵
factor for the flow of Bingham plastic fluids in smooth pipes by applying Adomian
Decomposition Method (ADM) and the more effective Restarted ADM (RADM) on the
Colebrooke Eq. [30] for turbulent flows and on the Buckingham Eq. [29] for laminar
flows. The explicit relations obtained from RADM were highly accurate for turbulent
flows (within 0.005% error) and fairly accurate for laminar flows (within 5.2% error). In
11
the laminar regime, errors were reported at high Reynolds (Re=1x105) and Hedstrom
numbers (He=1x1010).
4𝐾2
𝐾1 + 3
𝐾𝐾
(𝐾1 + 4 1 2 )
𝐾1 + 3𝐾2
𝑓= (19)
3𝐾2
1+ 4
𝐾𝐾
(𝐾1 + 4 1 2 )
𝐾1 + 3𝐾2
16 16𝐻𝑒
𝐾1 = + (20)
𝑅𝑒𝐵 6𝑅𝑒𝐵 2
16𝐻𝑒 4
𝐾2 = − (21)
3𝑅𝑒𝐵 8
expression for mean velocity of a Bingham plastic fluid in terms of yield stress and wall
shear stress.
𝐷𝜏𝐵 4 𝜏𝐵 1 𝜏𝐵 4
𝑢𝑚𝑒𝑎𝑛 = [1 − + ( ) ] (22)
2𝜇𝑝 3 𝜏𝑤 3 𝜏𝑤
For small values of 𝜏𝐵 /𝜏𝑤 , as is seen from the results of the loop test experiment,
𝐷 4
𝑢𝑚𝑒𝑎𝑛 = (𝜏𝑤 − 𝜏𝐵 ) (23)
8𝜇𝑝 3
By replacing 𝜏𝑤 in Eq. (23) with that from Eq. (6), the above equation can be re-
written and friction factor can be expressed in terms of values that may be experimentally
obtained i.e. Bingham yield stress, Bingham plastic viscosity and mean velocity.
12
5.6. Darby-Melson
R. Darby and J. Melson developed a friction factor correlation by introducing a
modified Reynolds number to the Buckingham-Reiner equation. The model was initially
developed for Bingham plastic fluids with 𝜏𝐵 < 2Pa and 𝜇𝑝 < 0.2 Pas by modifying the
16
𝑓= (25)
𝑅𝑒𝑚𝑜𝑑
6𝑅𝑒𝐵2
𝑅𝑒𝑚𝑜𝑑 = (26)
6𝑅𝑒𝐵 + 𝐻𝑒
largest root of the Buckingham-Reiner equation. They introduced two parameters 𝑃 and
𝐵 𝜏𝐵 𝐷
𝑃= ;𝐵 = (27)
𝑅𝑒𝐵 𝑢𝜇𝑝
2 𝐵
𝑄= ( + 2) (28)
𝑅𝑒𝐵 3
1 𝑃4 1 𝑃8 1 𝑃12
𝑓 = 4𝑄 − − − −⋯ (29)
12 𝑄 3 192 𝑄 7 9216 𝑄11
The first term on the right-hand side of Eq. (29) is analogous to that obtained by
dropping the fourth order term in Eq. (146. For small values of B, Eq. (30) is a good
1 8 9 𝐵4
𝑓= [16 + 𝐵 − ] ; 𝐵 = 𝑆𝑚𝑎𝑙𝑙 (30)
𝑅𝑒𝐵 3 32 (𝐵 + 6)3
2𝐵
𝑓= ; 𝐵 = 𝐿𝑎𝑟𝑔𝑒 (31)
𝑅𝑒𝐵
13
5.8. Hanks
The friction factor model developed by R.W. Hanks for the laminar flow of
Herschel-Bulkley fluids is shown below in Eq. (32). Eqs. (33) and (34) define the factor
𝜓 and dimensionless plug radius 𝜉0 . For Newtonian fluids, the friction factor is obtained
from Eq. (32) by taking 𝜓=1. For Bingham plastic fluids, 𝜓 and 𝜉0 are determined by
taking 𝑛=1. It is observed that when wall shear stress 𝜏𝑤 approaches the Bingham yield
stress 𝜏𝐵 , 𝜉0 →1 and 𝜓→0 resulting in infinitely high values of friction factor. In the
16
𝑓= (32)
𝜓𝑅𝑒𝐵
𝑛
𝑛 (1
(1 − 𝜉0 )2 2𝜉0 (1 − 𝜉0 ) 𝜉02
𝜓 = (1 + 3𝑛) − 𝜉0 )1+𝑛 [ + + ] (33)
(1 + 3𝑛) (1 + 2𝑛) (1 + 𝑛)
𝜏𝐵
𝜉0 = (34)
𝜏𝑤
6. Numerical Model
The numerical study of the backfill flow in the flow loop apparatus is performed
using the commercial computational fluid dynamics (CFD) solver ANSYS Fluent. The
dimensions from Fig. 2. Only the portion of the loop test that has pressure gauges is
considered for numerical solving. The geometry is meshed using ANSYS Meshing with
a boundary layer mesh set to the O-grid on the pipe circumference towards the centre
14
ensuring finer mesh elements in the boundary. Six meshes with different degrees of
refinement were produced and the case with highest Reynolds number, (with lowest mass
concentration) was used to run the grid independence study. The results of the study are
shown in Fig. 3. The mesh with 3 million elements was selected for the numerical model
Bingham properties from loop test are used to model the fluid behaviour. A Bingham
fluid is one that requires a finite shearing stress to initiate motion and for which there
exists a linear relationship between the shearing stress greater than the initiating and the
resulting velocity gradient [38]. This minimum stress required is called critical yield
stress and it causes the fluid to flow as a viscous material with a finite viscosity. The
𝜏 = 𝜏𝐵 + 𝜇 𝑃 𝛾 (35)
The apparent viscosity 𝑎𝑝𝑝 is obtained by dividing Eq. (35) by shear rate 𝛾
throughout.
𝜏 𝜏𝐵
𝑎𝑝𝑝 = = + 𝜇𝑃 (36)
𝛾 𝛾
For Bingham fluids, the apparent viscosity 𝑎𝑝𝑝 is infinite up until the applied
shear stress is below 𝜏𝐵 and beyond that it has a constant value 𝜇𝑃 . For numerical
modelling a small value of shear rate 0.001 is chosen for stitching up the discontinuity at
zero shear rate. During numerical solving, as the shear rate increases from 0 to 0.001,
𝑎𝑝𝑝 has a high value behaving almost as a solid. Beyond that, the Eq. (33) is obeyed.
15
6.3. Governing Equations
The numerical model solves the equations for mass and momentum conservation
∇ ∙ 𝜌𝑢
⃗ =0 (37)
preceding the inlet in the actual set up. To counter this, a 2.0 m developing length is
included before the first pressure gauge. The average flow velocity is known, and a
uniform velocity profile is specified at the inlet surface. The walls have been specified
with no-slip condition, and outlet with gauge pressure equal to zero. The operating density
is set equal to the backfill density to ensure no hydrostatic pressure variation. This
facilitates the removal of gravitational head in the pressure variation in the pipe for
solution of a two-dimensional axisymmetric pipe flow using the selected mesh density
for the two mines. The least squares cell-based gradient evaluation is used for its higher
accuracy and cheaper computation requirements over its node-based counterparts. Due to
the absence of strong body forces and stability issues, the second-order pressure scheme
is used. Different schemes for the pressure-velocity coupling and momentum were tested
to determine the most accurate velocity profile matching the analytical Bingham case.
The two pressure-velocity coupling algorithms tested were semi-implicit pressure linked
equation (SIMPLE) and pressure-implicit with splitting of operators (PISO). For spatial
16
discretization 1st order, 2nd order and 3rd order MUSCL (monotone upstream-centred
schemes for conservation laws) were tested. The velocity gradient from all the above
schemes were compared against the analytical results obtained from the below Eq. (40).
𝛥𝑃 2 𝑟 2
𝑢 = 𝜏𝐵 (𝑟 − 𝑅) + ( 𝑅 ) [1 − ( ) ] (40)
4𝜇𝑃 𝐿 𝑅
The results from the analyses are displayed below in Figs. 5a and 5b. The SIMPLE
2nd order produced the best agreement with the analytical solution for both mine samples.
Therefore, for all the remaining simulations were carried out using SIMPLE 2nd order
discretization scheme.
7. Test Results
7.1. Rheological Parameters from the Loop Test
The rheological parameters developed from the loop test experiment are presented
in Table 3. The experiment showed an increase in Bingham yield stress and Bingham
plastic viscosity with increasing solid concentrations. This phenomenon can be attributed
to the low mean particle and inter-particle distances in high solid concentration flows
resulting in strong interference effects that in-turn increase resistance to shearing [39].
The Bingham plastic viscosity determined through the loop test experiment is
input in Eq. (9) to get the Bingham Reynolds number and in Eq. (10) to get the
Generalized Bingham Reynolds number. A large deviation was seen between the two sets
of values. A parity plot between the two values displayed in Fig. 6 shows the
𝑅𝑒𝐺𝐸𝑁 𝐵𝐼𝑁𝐺𝐻𝐴𝑀 under predicts the Reynolds number in comparison to 𝑅𝑒𝐵 which results
in an over prediction of the friction factor. The friction factor at low velocities exceeded
1500 which is unrealistic. This confirms that the Generalized Bingham Reynolds number
17
proposed by K. Madlener, B. Frey and H.K. Ciezki is not suitable for high concentration
paste backfill. The Bingham Reynolds number in Eq. (9) is considered apt for this
Eight friction factor models from Eqs. (17) to (34) were compared against results
from CFD numerical simulation and loop test experiment conducted at three temperatures
16°C, 26°C and 35°C at low Reynolds numbers (<2300). The results are plotted in below
in Figs. 7a – 7f.
The Swamee & Aggarwal model as displayed in Eq. (18) showed the best
agreement with loop test data. The authors in their research noted less than 1% error in
their implicit formulation. The Jennifer Sinclair Curtis model as shown in Eq. (24) also
closely agrees with loop test data. Both models show better agreement with experimental
data at higher values of 𝑅𝑒𝐵 and start deviating marginally as 𝑅𝑒𝐵 reduces.
The Hagen-Poissuille model shown in Eq. (17) was initially derived for
Newtonian fluids and does not consider the Bingham yield stress. Despite not agreeing
with the loop test results, it is observed that this model agrees more closely than Hanks,
Ihle & Tamburrino, Danish & Kumar and Darby & Melson models in most cases putting
into question the validity of these models for high concentration paste backfill flow.
The Darby and Melson model as shown in Eq. (25) is very simple and was
originally developed for Bingham plastic fluids with very low yield stresses and plastic
viscosities (𝜏𝐵 < 2 Pa and 𝜇𝑝 < 0.2 Pas). The results from this model were very close to the
Danish & Kumar model as shown in Eq. (19). From their research, the authors concluded
their model had lower errors for turbulent flow (within 0.005%) in comparison to laminar
flows (within 5.2%). Both these models largely under predicts the friction factor.
18
The Ihle and Tamburrino model is divided into two cases; for small values of
Bingham number as shown in Eq. (30) and for large values of Bingham number as shown
in Eq. (31). In their study, the authors do not clearly prescribe what is classified as a small
or large Bingham number. From the loop test experiment, the Bingham number that were
recorded range from 6 to 117. Both models under predict the friction factor seen in the
loop test.
The Hanks model as shown in Eqs. (32) to (34) under predicts the friction factor.
The reason for this may be due to empirically obtaining the value of the factor 𝜓 and
dimensionless plug radius 𝜉0 which is a ratio of Bingham yield stress to wall shear stress.
The results from the CFD numerical simulation show very good agreement with
loop test data. The accuracy of the numerical simulation may further be improved by
accounting for the particle-particle interaction which is not incorporated in the current
model. For thick backfill slurries with high viscosity this phenomenon has a greater effect
at high Reynolds number. Another improvement can be done by accounting for the
temperature changes during the flow and how that may in turn affect the material
properties. The mine backfill which is in practice a multiphase flow (tailings and water)
In summary, the results from this study as seen in Figs. 7a – 7f can be broadly
classified into two groups. The Swamee & Aggarwal, Jennifer Sinclair Curtis and the
CFD models show good agreement with each other and are far off from Danish & Kumar,
Hagen-Poisseuille, Darby & Melson, Ihle & Tamburrino and Hanks models.
19
7.4. Statistical Analysis
For better interpretation of the vast amount of data generated from each friction
determination is a statistical measure of how close the data is to the fitted regression line.
It is the ratio of explained variation to total variation and lies between 0% and 100%.
Root mean square error (RMSE) is the standard deviation of residuals. It is a measure of
how spread out the predicted errors are. Mean absolute percent error (MAPE) is a measure
of the size of error in percentage terms. Percentage errors are scale-independent and range
variation of MAPE that ranges from -200 to 200. It overcomes the bias MAPE has
favouring predicted values that are lower than actual values. Two model selection criteria
were used in this study; Akaike Information Criterion (AIC) and Bayesian Information
Criterion (BIC). AIC and BIC are based on maximum likelihood estimates of the model
parameters. Table 4 below showcases the governing expression used in each statistical
model and its interpretation criterion. Tables 5 and 6 showcase the results obtained from
The coefficient of determination R2 gives high values above 0.97 for CFD,
Swamee & Aggarwal, Danish & Kumar, Jennifer Sinclair Curtis, Darby & Melson and
Ihle & Tamburino B=Large. The RMSE analysis shows lowest regression for Swamee &
Aggarwal (0.647) followed by CFD (0.685) and Jennifer Sinclair Curtis (0.966) while all
other models have values above 2.5. A similar trend is seen from MAPE and SMAPE
analyses where Swamee & Aggarwal has the lowest percentage error (3.984 and 3.860)
followed by Jennifer Sinclair Curtis and CFD. The model selection criteria AIC and BIC
gives the lowest values for Swamee & Aggarwal (116.810 and 116.486) followed by CFD
and Jennifer Sinclair Curtis while all other models have values above 200.
20
From the statistical analyses performed it can be concluded that Swamee &
Aggarwal model has the best fit with the experimental results obtained from the loop test.
The CFD model has the second-best fit based on RMSE, AIC and BIC while Jennifer
Sinclair Curtis has a better fit based on MAPE and SMAPE. It is important to note that
the R2 model shows inconsistent results by having high values for Danish & Kumar,
Darby & Melson and Ihle & Tamburrino B=Large which neither correlates with the
results from the other statistical models nor with the results in Figs. 7a – 7f. This indicates
more than three kilometres was measured using a PSI pill [40] in the mine. The pill was
introduced into the paste backfill pipeline at the inlet and collected at the discharge. The
data from the pill was extracted and compared against the frictional pressure loss
determined using the Swamee & Aggarwal friction factor model. The gravitational
pressure was removed from the PSI pill and Swamee & Aggarwal analyses using the
Fig. 8a and the results were included in the comparison. Fig. 8b below shows the results
of this comparison and a good agreement is observed between the in-situ pill test results
numerical model and Swamee & Aggarwal friction factor model within an error
percentage of 5%. The numerical model showed better agreement towards the end of the
pipeline than the start. This comparison confirms the accuracy of the Swamee &
Aggarwal model in predicting the friction factor and the pressure losses in long distance
21
adopted with the gravitational pressure decoupled, and the pressure loss is purely
frictional, which show the accuracy and validity of both numerical and friction factor
8. Conclusion
An experimental setup, called the loop test, was used to determine the pressure
gradients at different flow velocities for the flow of high concentration (67-75% by mass)
paste backfill samples prepared from tailings from two mines in Canada. The flow lies in
the laminar regime with Reynolds number < 2300. The friction factors for the flow were
obtained from the loop test experiment, numerical simulations and eight friction factor
correlations.
The Generalized Bingham Reynolds number was shown to under predict the
Reynolds number in comparison to loop test data and over predict the friction factor. It is
deemed incompatible for high concentration paste backfill flows. The Bingham Reynolds
Swamee & Aggarwal friction factor correlation was seen to best predict the
friction factor at the Reynolds number and temperature ranges tested in the loop test
experiment. The second closest agreement was recorded with the Jennifer Sinclair Curtis
model. The models proposed by Hagen-Poiseuille, Danish & Kumar, Darby & Melson,
Ihle & Tamburrino and Hanks predominantly under predict the friction factor. The results
from six statistical analyses confirm that Swamee & Aggarwal correlation best predicts
the friction factor from the experiment closely followed by Jennifer Sinclair Curtis model.
The CFD modelling of the flow loop system was done by considering the paste backfill
results show very good agreement with the loop test data. The CFD model can be further
22
improved by accounting for the particle-particle interaction and temperature dependence
compared against the Swamee & Aggarwal correlation showed good agreement between
the two. Future work should be targeted at testing a wide range of mine tailings with
encompasses all paste backfill will be useful for engineers, designers and researchers in
This study shows that care should be taken in selecting a friction factor correlation
for non-Newtonian fluids. There exist numerous relationships derived for specific flow
regimes and fluid properties that remain valid only for the confines of the parameters
Future work will focus on the two-fluid CFD model to capture the effects of
individual phases of water and tailings and their interactions. The current work will serve
as a pre-cursor to the two-fluid model which will account for particle treatment and
Acknowledgement
The authors acknowledge the support of the Natural Sciences and Engineering
program. We also gratefully acknowledge Paterson & Cooke for the use of their loop test
facility to perform the test work. The third author acknowledges Mitacs Globalink
23
Nomenclature
𝜌 Density kg/m3
𝑃 Pressure Pa
𝐿 Unit length m
𝐿𝑉 Vertical length m
𝑆𝑔 Specific gravity -
𝛥𝑃 Pressure loss Pa
𝜏 Shear stress Pa
24
𝜏𝑤 Wall shear stress Pa
𝑢
⃗ Velocity vector m/s
𝐷 Pipe diameter m
𝑟, 𝑅 Pipe radius m
𝐻𝑒 Hedstrom number -
𝑓 Friction factor -
𝐾1 , 𝐾2 Constants -
𝐵 Bingham number -
25
𝜉0 Dimensionless plug radius -
𝐼 Identity matrix -
𝑁 Sample size -
𝑝 Number of parameters -
References
[1] J. Barrett, M. Coulthard, P. Dight, Determination of fill stability, in: 12th Can.
Rock Mech. Symp., Canadian Institute of Mining, Metallurgy and Petroleum,
Sudbury, Ontario, 1978: pp. 85–91.
[2] B.G. Lottermoser, Recycling, Reuse and Rehabilitation of Mine Wastes, Elements.
7 (2011) 405 LP-410.
http://elements.geoscienceworld.org/content/7/6/405.abstract.
[4] D. Wu, Y. BG, L. YC, Pressure drop in loop pipe flow of fresh cemented coal
gangue-fly ash slurry: Experiment and simulation, Adv. Powder Technol. 26
(2015) 920–927. doi:10.1016/j.apt.2015.03.009.
[5] N. Cruz, Y. Peng, Rheology measurements for flotation slurries with high clay
contents – A critical review, Miner. Eng. 98 (2016) 137–150.
doi:https://doi.org/10.1016/j.mineng.2016.08.011.
26
[6] N. Gharib, B. Bharathan, L. Amiri, M. McGuinness, F.P. Hassani, A.P. Sasmito,
Flow characteristics and wear prediction of Herschel-Bulkley non-Newtonian
paste backfill in pipe elbows, Can. J. Chem. Eng. 95 (2017) 1181–1191.
doi:10.1002/cjce.22749.
[7] A. Rawat, S.N. Singh, V. Seshadri, Erosion wear studies on high concentration fly
ash slurries, Wear. 378–379 (2017) 114–125.
doi:https://doi.org/10.1016/j.wear.2017.02.039.
[8] D. Wu, M. Fall, S.J. Cai, Coupling temperature, cement hydration and rheological
behaviour of fresh cemented paste backfill, Miner. Eng. 42 (2013) 76–87.
doi:https://doi.org/10.1016/j.mineng.2012.11.011.
[9] K.J. Creber, M.F. Kermani, M. McGuinness, F.P. Hassani, In situ investigation of
mine backfill distribution system wear rates in Canadian mines, Int. J. Mining,
Reclam. Environ. 31 (2017) 426–438. doi:10.1080/17480930.2017.1339169.
[10] P.K. Senapati, B.K. Mishra, Design considerations for hydraulic backfilling with
coal combustion products (CCPs) at high solids concentrations, Powder Technol.
229 (2012) 119–125. doi:https://doi.org/10.1016/j.powtec.2012.06.018.
[11] J. Liu, R. Wang, F. Gao, J. Zhou, K. Cen, Rheology and thixotropic properties of
slurry fuel prepared using municipal wastewater sludge and coal, Chem. Eng. Sci.
76 (2012) 1–8. doi:https://doi.org/10.1016/j.ces.2012.04.010.
[12] K.M. Assefa, D. Kaushal, A comparative study of friction factor correlations for
high concentrate slurry flow in smooth pipes, J. Hydrol. Hydromechanics. 63
(2015) 13. doi:10.1515/johh-2015-0008.
[13] S. Chandel, V. Seshadri, S.N. Singh, Effect of Additive on Pressure Drop and
Rheological Characteristics of Fly Ash Slurry at High Concentration, Part. Sci.
Technol. 27 (2009) 271–284. doi:10.1080/02726350902922036.
[14] S. Chandel, S.N. Singh, V. Seshadri, Transportation of high concentration coal ash
slurries through pipelines, Part. Sci. Technol. 1 (2010) 1–9.
[15] K.C. Wilson, A.D. Thomas, A new analysis of the turbulent flow of non-newtonian
fluids, Can. J. Chem. Eng. 63 (1985) 539–546. doi:10.1002/cjce.5450630403.
[16] F.A. Morrison, Data correlation for friction factor in smooth pipes, Houghton,
Michigan, 2013.
[17] D. Wu, B. Yang, Y. Liu, Transportability and pressure drop of fresh cemented coal
gangue-fly ash backfill (CGFB) slurry in pipe loop, Powder Technol. 284 (2015)
218–224. doi:https://doi.org/10.1016/j.powtec.2015.06.072.
[18] P.K. Swamee, N. Aggarwal, Explicit equations for laminar flow of Bingham
plastic fluids, J. Pet. Sci. Eng. 76 (2011) 178–184.
doi:https://doi.org/10.1016/j.petrol.2011.01.015.
[19] C. Qi, Q. Chen, A. Fourie, J. Zhao, Q. Zhang, Pressure drop in pipe flow of
cemented paste backfill: Experimental and modeling study, Powder Technol.
27
(2018). doi:https://doi.org/10.1016/j.powtec.2018.03.070.
[20] K. Madlener, B. Frey, H.K. Ciezki, Generalized reynolds number for non-
newtonian fluids, EUCASS Proc. Ser. – Adv. Aerosp. Sci. 1 (2009) 237–250.
https://doi.org/10.1051/eucass/200901237.
[22] C.W. Bakker, C.J. Meyer, D.A. Deglon, Numerical modelling of non-Newtonian
slurry in a mechanical flotation cell, Miner. Eng. 22 (2009) 944–950.
doi:https://doi.org/10.1016/j.mineng.2009.03.016.
[24] F. Jiang, P. Zhao, G. Qi, N. Li, Y. Bian, H. Li, T. Jiang, X. Li, C. Yu, Pressure
drop in horizontal multi-tube liquid–solid circulating fluidized bed, Powder
Technol. (2018). doi:https://doi.org/10.1016/j.powtec.2018.04.003.
[26] E.J. Garcia, J.F. Steffe, Comparison of friction factor equations for non-Newtonian
fluids in pipe flow, J. Food Process Eng. 9 (1986) 93–120. doi:10.1111/j.1745-
4530.1986.tb00120.x.
[27] S.P. Sutera, R. Skalak, The History of Poiseuille’s Law, Annu. Rev. Fluid Mech.
25 (1993) 1–20. doi:10.1146/annurev.fl.25.010193.000245.
[29] E. Buckingham, On plastic flow through capillary tubes, in: ASTM, 1921: pp.
1154–1161.
[30] C.F. Colebrook, Turbulent flow in pipes, with particular reference to the transition
region between the smooth and pipe laws, J. Inst. Civ. Eng. 11 (1939) 133–156.
doi:10.1680/ijoti.1939.13150.
[31] L. Prandtl, The mechanics of viscous fluids, Durand, W.F. Aerodyn. Theory. 3
(1935) 34–208.
[32] R. Darby, J. Melson, How to predict the friction factor for flow of Bingham
plastics, Chem. Eng. 88 (1981) 59–61.
28
[33] P.T. Slatter, The turbulent flow of non-Newtonian slurries in pipe, in: 8th Conf.
Transp. Sediment. Solid Part., Prague, 1995.
[35] J.S. Curtis, Slurry Transport, in: M.J. Rhodes (Ed.), Introd. to Part. Technol., 2nd
ed., 2008: pp. 91–116. doi:10.1002/9780470727102.
[36] C.F. Ihle, A. Tamburrino, A note on the Buckingham equation, Can. J. Chem. Eng.
90 (2012) 944–945. doi:10.1002/cjce.20606.
[38] G.W. Govier, K. Aziz, The flow of complex mixtures in pipes, Van Nostrand
Reinhold, 1972.
[39] S. Chandel, S.N. Singh, V. Seshadri, Deposition characteristics of coal ash slurries
at higher concentrations, Adv. Powder Technol. 20 (2009) 383–389.
doi:https://doi.org/10.1016/j.apt.2009.06.004.
[40] Camiro. (2004). "Know Your Pipeline - PSI Pill (Parr Innovations) " Retrieved
August 30, 2011, from http://www.camiro.org/Mining_Products.htm
29
Table 1
Mine-A and B tailings composition.
30
Table 2
Mine-A and B tailings characterization.
d80 60 µm 20 µm
d50 25 µm 10 µm
d20 8 µm 7 µm
% < 20 µm 45 % 80 %
31
Table 3
Rheological parameters developed from loop test data for Mine-A and Mine-B.
Mine-A Mine-B
32
Table 4
Governing expression of each statistical model and its interpretations.
Coefficient of ∑𝑁 ̅ ̂ 2
𝑖=1(𝐹𝑡 − 𝐹𝑡 )
The closer the value is to 1,
2
𝑅 = 𝑁
determination R2 ∑𝑖=1(𝐹𝑡 − 𝐹̅𝑡 )2 the better the model
𝑅𝑆𝑆
Akaike Information 𝐴𝐼𝐶 = 𝑁 + 𝑁𝑙𝑜𝑔(2𝜋) + 𝑁𝑙𝑜𝑔 ( ) The smaller the value, the
𝑁
Criterion (AIC) better the model
+ 2(𝑝 + 1)
𝑅𝑆𝑆
Bayesian Information 𝐵𝐼𝐶 = 𝑁 + 𝑁𝑙𝑜𝑔(2𝜋) + 𝑁𝑙𝑜𝑔 ( ) The smaller the value, the
𝑁
Criterion (BIC) better the model
+ (𝑙𝑜𝑔𝑁)(𝑝 + 1)
33
Table 5
Statistical analysis of friction factor models against loop test experiment for Mine-A.
34
Table 6
Statistical analysis of friction factor models against loop test experiment for Mine-B.
35
Figures
Fig. 1
Fig. 2
36
Fig. 3
Fig. 4a
37
Fig. 4b
Fig. 5a
38
Fig. 5b
Fig. 6
39
Fig. 7a
Fig. 7b
40
Fig. 7c
Fig. 7d
41
Fig. 7e
Fig. 7f
42
Fig. 8a
Fig. 8b
43
Figure Captions
Fig. 1. Mine-A and B particle size distribution.
Fig. 2. Schematic arrangement of the flow loop test experimental apparatus (All
dimensions are in mm).
Fig. 3. Grid independence study.
Fig. 4. Loop geometry used in CFD model; (a) computational domain and (b) mesh
profile.
Fig. 5. Numerical model verification against various discretization schemes for (a) Mine-
A; and (b) Mine-B.
Fig. 6. Bingham Reynolds number – Generalized Bingham Reynolds number parity plot.
Fig. 7. Comparison of friction factor models for Mine-A at (a) 16°C; (b) 26°C; (c) 35°C;
and for Mine-B (d) 16°C; (e) 26°C; and (f) 35°C.
Fig. 8. (a) Schematic diagram of the mine-field pipeline used in CFD model; and (b)
Comparison of pressure reading between in-situ mine data, Swamee & Aggarwal friction
factor model and CFD model.
44