Building and Environment 121 (2017) 200e214

Contents lists available at ScienceDirect

Building and Environment

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

Correlating variability of the leakage characteristics with the hydraulic

performance of an auxiliary ventilation system
Saad Akhtar, Mustafa Kumral, Agus P. Sasmito*
Department of Mining and Materials Engineering, McGill University, 3450 University Street, Montreal QC H3A0E8, Canada

a r t i c l e i n f o a b s t r a c t

Article history: Ventilation is one of the key factors in controlling underground working environment by providing
Received 24 February 2017 sufficient amount of fresh air for breathing, dispersing harmful gasses and dust and to some extent for
Received in revised form heating/cooling. Insufficient airflow is dangerous for the working face and can lead to fatalities. Duct
16 May 2017
leakage is the most common reason for the insufficient fresh air in underground working and has been
Accepted 19 May 2017
the subject of many studies in the literature. However, the main focus has been on ascertaining its impact
Available online 20 May 2017
on the ventilation requirements of the underground environment. This study aims to identify key var-
iables associated with duct leakage that significantly impacts the power consumption levels of auxiliary
Keywords:
Auxiliary ventilation
fans which form an integral part of the underground ventilation system. A three-dimensional Compu-
CFD tational Fluid Dynamics (CFD) modeling approach is undertaken in conjunction with Monte Carlo sim-
Duct leakage ulations and multiple regression analysis to quantify the effect of duct leakage on the fan operating point
Fan operating point and discharge flow rate towards the working face. Various cases involving the positioning, orientation,
Flow resistance and size of the rupture in the ventilation duct are simulated, and their respective effects on fan operating
point and power levels are determined. Results indicate that the operating point of a fan for ventilation
ducts is strongly correlated with the position and size of the rupture, resulting in reduced delivery of
ventilation air towards the working face for different levels of fan power consumption.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction
Sufficient fresh airflow is the key to maintaining the good un-
derground working environment as it is used for breathing,
dispersing harmful gasses and dust, and thermal comfort. Insuffi-
cient fresh air can lead fatalities due to oxygen deficiency, explosion
due to flammable gas (such as methane in a coal mine), health
problems due to dust and diesel emission. Duct leakage has been
the most common reason for the insufficient fresh air. Accurately
incorporating the effect of duct leakage in the design process of the
auxiliary ventilation system has been the Achilles heel for venti-
lation engineers [1e4]. Meticulous consideration of the impact of
flow leakage from auxiliary ventilation duct is very important in
ensuring the optimum levels of fresh air delivered while main-
taining acceptable levels of contaminants in the underground
environment [5e8]. One direct corollary of this consideration is a
selection of appropriate fan for the job. Fan's operating point, which
is defined by its operating pressure and corresponding flow rate
* Corresponding author.
E-mail address: agus.sasmito@mcgill.ca (A.P. Sasmito).
delivered, is dependent upon the level of air leakage from ducts
which in turn is influenced by different factors such as the type of
duct material, rupture in ducts, duct joints etc. [9e11]. Hence in
order to accomplish a safe and efficient ventilation system, it is
pertinent to examine the mechanics of flow leakage in detail [12].
The quantification of air leakage and its effect on the overall
energy consumption has been a subject of several experimental and
numerical studies. Kalamees et al. [13] and Belleudy et al. [14]
conducted experimental and numerical studies, exploring a
related issue, where they quantified the average leakage rates from
the structures. In another study by Jokisalo et al. [15] the effect of
infiltration on the energy consumption of a building under cold
environment is evaluated. The study found out that infiltration can
cause a spike of up to 30% in the overall energy consumption by
space heating equipment.
There are numerous studies in the literature which have
attempted to incorporate the effect of leakage while analyzing the
performance of auxiliary ventilation systems. Lee et al. [16] inves-
tigated the impact of leakage through airways to analyze ventila-
tion efficiency of coal mines. They presented a mathematical
formulation to determine the pressure drop and air flow rate while

http://dx.doi.org/10.1016/j.buildenv.2017.05.029
0360-1323/© 2017 Elsevier Ltd. All rights reserved.
 S. Akhtar et al. / Building and Environment 121 (2017) 200e214
incorporating air leakage in ducts. Wolski et al. [17] presented a
numerical study to determine leakage factors in ducts in order to
estimate system performance for different fan configurations. The
study came up with an optimized system of series of fans in an
auxiliary ventilation system. These early studies though valuable
needed further model refinement in order to effectively incorpo-
rate the effect of duct leakage on overall ventilation efficiency.
Auld [9] studied the difference between auxiliary ventilation
characteristics for leak-free and leaky ducts. The study introduced
the concept of duct efficiency which takes into the account the
variability in duct leakage with the increasing fan operating pres-
sure. It uses the spreadsheet method to compute the fan operating
points in a leaking duct. The study concluded that it would be faulty
to select fan based on a leak-free system characteristic curve.
Furthermore, it posited that the leakage condition of the duct
become increasingly important as the duct length increases. In
another study, Onder et al. [18] attempted to optimize the opera-
tional factors affecting the volumetric flow rate of air reaching the
working face. Factors considered in the study include duct diam-
eter, friction factor, air flow rate at the duct inlet, duct length and
resistance coefficient of a leakage path. The study developed a
computer program to solve for the fan operating point and air flow
rate at working face while incorporating Kirchoff resistance anal-
ogy to model duct leakage. It was found that duct diameter and fan
flow rate significantly affect the ventilation output. The volume
flow rate to the working face was less sensitive to duct length,
friction factor and resistance coefficient for duct leakage.
In another study, Onder et al. [19] presented statistical model of
the volume flow rate reaching the working face using the field data
from an underground mine. Multiple regression analyses were
carried out to study the effect of variables such as duct diameter,
length, friction factor, resistance coefficient of leakage path, fan
operating pressure and flow output etc. on the volumetric flow rate
at working face. Statistically, significant correlations were devel-
oped with acceptable accuracy to carry out optimization studies for
ventilation design.
All of the afore-mentioned studies lay their prime focus on
ascertaining the overall impact of leakage on the ventilation system
characteristics. However, a component-level approach of tackling
the leakage issue lacks in the literature. Variables such as rupture
positioning, leak orientation, and rupture size are some of the key
considerations that must be taken into account while designing
such systems. These factors are of paramount importance, espe-
cially in the underground mining industry, as in some countries, the
flow rate in working space is regulated by law, e.g. in typical Ca-
nadian mines in Ontario province requires 0.063 m3/s of fresh air
per kW power of diesel machine. Thus, if the airflow rate at the
working face is below the minimum requirement due to leakage,
the mine operation can be shut down by law. Also, it is not un-
common in underground operations to have tears and ruptures in
the auxiliary ventilation ducts due to the movement of heavy ma-
chinery below the ducts. This leads to air leakage and inefficiency of
the ventilation system. On top of that, insufficient airflow at the
working face can lead to fatalities, especially in the underground
coal mines that emit flammable gasses such as methane. This study
aims to fill that gap by studying the effect of these parameters on
fan power requirement and system volumetric flow rate at the
working face through well-established computational fluid dy-
namics (CFD) approach coupled with Monte-Carlo and regression
models. Furthermore, as pointed in the literature [20], the resis-
tance to air flow in a ventilation system is critical to the overall
performance of the system. Thus, quantification of the impact of
leakage on system's efficiency is crucial. Note that the leakage
phenomenon is complex in nature, highly non-linear and is
affected by many factors that interacts to each other. Thus far,
201
however, none has attempted to quantify the effect of these pa-
rameters from mechanistic and continuum point of view. There-
fore, for first approximation, the scope of present study is limited to
three leakage factors, i.e. rapture location, rapture size and rapture
orientation. The effect of other parameters, such as duct material,
duct joint, duct diameter, duct length and so forth will be evaluated
in our future study.
The geometrical parameters associated with the rupture are
assumed to be stochastic in nature. Consequently, their impact on
the output variables is also studied probabilistically. Monte Carlo
statistical simulations have been carried out in conjunction with
regression analysis to determine the dependency of rupture char-
acteristics on response variables. Based on the analysis of the dis-
tribution of the output variables, recommendations are drawn for
estimating the likelihood of instances which are critical to the fan
performance and ventilation requirements. Resistance analogy for
the duct leakage is also developed stochastically to create an easy to
follow the method in the context of ventilation professionals.
2. Methodology
This section provides a description of various steps involved in
the computational fluid dynamics (CFD) analysis coupled with
statistical methods. The flowchart showed in Fig. 1 elucidates the
interaction between CFD calculations and statistical methods for
this study. The rupture characteristics (size, location, and orienta-
tion) are taken to be random variables and are treated as such.
Monte Carlo Simulation (MCS) is used to generate a case list of
random samples for both independent and combined analysis of
the impact of input parameters on output variables. These random
samples are fed into the CFD simulation package which then yields
the case specific values of output variables such as fan power, fan
flow rate, leakage through rupture, flow rate towards the working
face and operating pressure of the fan. These results are then
analyzed by developing probability distributions for various
response variables which give important insights on ventilation
system design.
The preceding sub-section discusses the methodology for the
CFD study which includes flow model and boundary conditions
followed by a description of numerics. The last sub-section ex-
pounds the statistical method as tersely outlined afore.
2.1. Computational fluid dynamics (CFD) study
The schematic of the portion of the underground working face
modeled along with the rupture in the auxiliary duct is shown in
Fig. 2. The schematic shown in the figure comprises of the main
drift with a branching tunnel towards working face. It also contains
an auxiliary ventilation system as highlighted by the smaller curved
duct inside the outer drift structure. The auxiliary duct is taken to
be 68.58 m (225 ft) long with a 90
bend. The end of the duct is kept
at 15.24 m from the working face. The topology of the rupture and
the dimensions of the working face are also shown in Fig. 2.
2.1.1. Mathematical model
The objective of the CFD study is to solve for the flow field for
three-dimensional fluid flow taking into account ruptures and
bends in the auxiliary duct. To enhance the accuracy of the analysis,
surface roughness has also been incorporated into the model. For
the full-scale three-dimensional structure, the resolution of the
turbulent flow field requires solving continuity and momentum
conservation equations. The mathematical equations are presented
here in tensor notation for the Cartesian coordinate system. Equa-
tion (1) is the mathematical representation of the principle of
conservation of mass and is most commonly referred to as
202 S. Akhtar et al. / Building and Environment 121 (2017) 200e214

Fig. 1. Flow chart showing the interaction between statistical and Computational Fluid Dynamics portions of the study.

continuity. Equation (2) models the rate of conservation of mo- and is assumed to as Newtonian and viscous. Normal and shear
mentum principle for this problem. Air is taken as the working fluid stresses in Equation (2) are modeled by invoking stokes assumption
 S. Akhtar et al. / Building and Environment 121 (2017) 200e214 203

Fig. 2. Geometric details of the underground drift and rupture morphology (Schematic not to scale). The Roman numerals are represented as follows: I- Drift inlet, II- Drift outlet,
III-Working face, IV- Fan/auxiliary duct inlet, V-Rupture, and VI-Auxiliary duct outlet. Fixed properties: Flow rate through I and surface roughness of the walls. Variables: Flow rate
and pressure drop through IV, V and VI.

[21]. Stokes assumption takes the coefficient of bulk viscosity to be delivered at drift inlet.
zero which significantly simplifies the momentum equations. The Using the predefined value of roughness constant, roughness
assumption holds true for the majority of the ideal gasses. The height was varied in the model such that the required pressure
constitutive relation for deviatoric stress tensor is given by equa- drop was achieved in the simulation. This ‘tuned’ value of rough-
tion (3). The major unknown parameters in equations (1)e(3) are ness height comes out to be 0.2 m for the underground walls.
velocity field ‘u’ and pressure ‘p’. For the detailed description of
other variables used, the reader is referred to nomenclature section.
2.1.2. Boundary conditions
vðrui Þ Boundary labels are shown in Fig. 2 whereas corresponding
¼0 (1) values of flow rates, turbulence characteristics and surface rough-
vxi
ness for the ventilation system are provided in Table 1. Inlet flow-

 rate for the underground tunnel is taken to be 11.8 m3/s (25000
v rui uj vðpÞ v

¼ þ s (2) CFM) which is the standard value for an average Canadian mine
vxi vxj vxi ij with 187 kW (250 hp) of diesel machinery operating inside the
tunnel. This is assuming that the air flow requirement is 0.063 m3/s
!
2 vu vui vuj per kW (100 CFM/hp). Since only a single main drift is modeled, the
sij ¼  ðm þ mturb Þ k dij þ ðm þ mturb Þ þ (3) exit of the drift is taken to be at atmospheric pressure.
3 vxk vxj vxi
Inputs to the fan boundary condition type are the pressure-flow
The current study uses Reynolds Averaging approach to model rate polynomial equation where pressure is modeled as:
the turbulence stresses in the rate of momentum conservation
equation. Reynolds averaging has proved to be cost effective while
X
N
Dp ¼ Cn Q i1 (5)
providing acceptable levels of accuracy for underground ventilation i¼1
simulations [22,23] amongst other wide range of engineering ap-
plications [24]. The turbulence model used in this study is the The range of flow rate and corresponding total pressure drop
standard k-epsilon with non-equilibrium near wall formulation. across the fan depends on the type of fan selected. A plot of flow
According to several previous studies [25e28], k-epsilon provides rate versus pressure drop based on the correlation given in equa-
better local velocity results for the underground ventilation system. tion (5) is known as fan curve. For this study, quadratic polynomial
(i ¼ 3) is used to model the fan curve. The prime consideration for
2.1.1.1. Calibration of the surface roughness of the walls. In order to appropriate fan selection is efficiency which is a strong function of
accurately incorporate the effect of surface roughness, the CFD its operating point. The operating point, and thus efficiency, of a fan
model requires inputs in the form of roughness height and is determined by the intersecting point of fan curve and system
roughness constant. The roughness height takes into account the characteristic curve on flow rate versus pressure drop plot. For the
extent of the roughness of the wall whereas roughness constant graphical description of the concepts above, the reader is referred
represents the distribution of roughness over the whole wall. The to the work of Munson, Young, and Okiishi [29].
afore-mentioned wall roughness parameters are not reported in The system characteristic curve geometrically represents the
the literature.
For roughness constant, a value of 0.5 was assumed which sig- Table 1
nifies an even distribution of the roughness over the whole wall. Values of different system and flow parameters.
The roughness height was calculated through a verification
Parameter Units Values
method. Using the friction factor value of 0.019 for a moderately
Velocity at inlet m/s 0.54
rough surface [4] in equation (4) along with base case values for all
Velocity at duct inlet (for system characteristic curve) m/s 1.6e11.2
other parameters, the pressure drop was calculated. Surface roughness of the wall m 0.2
Roughness constant e 0.75
lp 2 Pressure outlet Pa 0
p ¼ kL Q (4)
A3 Turbulence intensity at drift Inlet % 5
Turbulence intensity at Duct Inlet % 5
here k, L, lp and A are friction factor, length, perimeter and area of Hydraulic diameter of drift m 6.013
Hydraulic diameter of Auxiliary duct m 1.07
the cross-section for the tunnel, respectively. Q is the flowrate
204 S. Akhtar et al. / Building and Environment 121 (2017) 200e214
relationship between the system head (pressure drop) and flow
rate. The characteristic curve for the auxiliary ventilation system is
obtained by computing pressure drop across the auxiliary ventila-
tion duct for several volumetric flow rates varying in the range of
2.36e16.52 m3/s. Based on the obtained system characteristic
curve, the fan curve for the model 3800-VAX-2100 with the blade
angle of 25
and impeller diameter of 38 in. was selected initially.
With this combination of fan curve and characteristic curve, how-
ever, the fan operating point lies in the 59% fan efficiency zone.
Thus, another fan with favorable characteristic curve was tested.
The 4200-VAX-2100 fan with 42 in impeller diameter operating at
980 RPM yields a more favorable operating point with a fan effi-
ciency of 73.5%. The pressure jump versus velocity relation for the
selected fan is given by:
 
Dp ¼  6:9  109 Q 2  ð0:00033ÞQ  2
The above equation is provided in CFD software as the fan
boundary condition for the inlet of the auxiliary duct. The same fan
characteristic is used for all the cases of rupture during the CFD
analysis.
2.1.3. Numerics
The CFD analysis for this study is performed using the fluid flow
toolbox in ANSYS Workbench 16.1. Geometry is developed and
meshed in ANSYS design modeler and Mesh module respectively,
whereas the governing equations of turbulent flow in the under-
ground are discretized and solved using FLUENT 16.1. The govern-
ing equation of the three-dimensional fluid flow are elliptic in
nature and are computed using segregated 3-D double precision
parallel computing solver. In order to discretize the continuity,
momentum, and turbulence equations, second order upwind
scheme is used. The coupling of pressure and velocity from mo-
mentum and continuity equation is solved using Semi-Implicit-
Method-for-Pressure-Linked-Equation (SIMPLE) algorithm.
Before proceeding with the CFD analysis, grid independency test
has been carried out to mitigate the impact of numerical biases on
the result. Fig. 3 shows the result of the grid independency test for
12 different mesh sizes ranging from 0.24 million to 4.9 million
elements. The mesh sensitivity analysis has been performed for the
leak-free auxiliary ventilation system. Also, the inlet of auxiliary
ventilation duct is taken as velocity inlet with a volumetric flow
rate of 2.359 m3/s (5000 CFM). Static pressure at the inlet is used as
Fig. 3. Mesh sensitivity study for the computational domain.
the output variable for mesh insensitivity test. Results from Fig. 3
show that mesh size of around 2.4 million strikes a fitting bal-
ance between mesh insensitivity and computational time. Hence a
mesh size of 2.4 million is used for all further flow simulations for
underground ventilation.
A total of 304, 3-dimensional turbulent flow computations were
carried out using the CFD software. Owing to a large number of
cases to be computed, a python script was written to automate the
repetitive tasks.
2.2. Monte Carlo Simulation and regression analysis
Monte-Carlo simulation is an algorithm based on generating
random samples to calculate numeric possibilities such that risk
analysis is implemented. The first step towards performing Monte
(6) Carlo Simulation (MCS) is to fit a probability distribution to un-
certain parameters. Rupture location (X1), size (X2) and orientation
(X3) on the auxiliary duct are uncertain or random variables whose
impact on volumetric flow rate at working face and fan operating
point is studied. The distributions are fitted using historical,
experience-based or experimental information. For X1, the study
assumes that probability of rupture over any length of the duct is
same. Hence a uniform distribution is fitted for X1 with a minimum
value of 0 m and the maximum value of 68.58 m. The range for X1
commensurates with the total length of the auxiliary duct which
was chosen in accordance with current mining industry norms.
Rupture size (X2) corresponds to the width of the rupture for fixed
length of the size of semi-circle around the duct. A normal distri-
bution is assumed for the rupture size with a mean value of rupture
width assumed to be 0.335 m (1.1 ft) with a standard deviation of
0.167 m (0.55 ft). For specified values of means and standard de-
viation the range of the rupture, width comes out to be between
0.08 m and 0.6 m. 0.6 m rupture width was chosen as the upper
limit since it corresponds to about 70e80% leakage in the duct
according to preliminary computations. The average value of
0.335 m was taken because CFD simulations with rupture size of
0.335 m corresponded to around 13e17% leakage by volume which
is an average value for most duct leakages in ventilation systems
[30]. As for the rupture orientation, only four distinct possibilities
are assumed i.e. rupture at the top half, bottom half, inner half
(facing the tunnel) and an outer half (facing the wall) of the duct.
Since X1, X2 and X3 are mutually independent, the correlation factor
in the MCS is taken to be zero.
After assuming the probability distributions and associated
distribution parameters of the random variables, MCS simulation is
run for a sample size of 100 cases first. It is important to note here
that two different scenarios are used to ascertain the impact of
stochastic predictors on response variables. One scenario (referred
to as scenario A from here on) comprises of varying one random
variable 100 times while keeping the other 2 constants (equal to
their average values). This generates a total of 204 cases for CFD
software to compute. 100 cases each are generated for X1 and X2
since they have a continuous probability distribution function.
Since the orientation of rupture is assumed to be discrete in nature
and has only 4 distinct possibilities, X3 constitutes of only 4 cases.
The second scenario (referred to as Scenario B from here on) of MCS
comprises of varying all the parameters simultaneously in order to
analyze the combined effect of all random variables. This generates
another 100 cases, thus taking the total CFD cases to be computed,
equal to 304.
After determining all the input cases, transport equations of
mass, momentum, and turbulence are numerically computed
which yield the case specific values of response variables. The
important response variables (Yi) analyzed in this study are fan
power (Y1), the flow rate through the fan (Y2), the flow rate through
 S. Akhtar et al. / Building and Environment 121 (2017) 200e214 205

rupture (Y3), the flow rate from discharge end (Y4), Percentage
leakage (Y5) and total fan pressure (Y6). The output results are again
probabilistically analyzed through MCS by developing probability
distribution functions of the response variables.
In order to get a good level of goodness of fit for response
probability distributions, the sample size of MCS was required to be
increased. More CFD simulations was an unrealistic option
considering that the sample size was to be increased as much as
10,000 points. Instead, another statistical tool was utilized to
improve the accuracy of the fit of distributions without compro-
mising on underlying physics.
2.2.1. Multiple regression analysis
Linear regression analysis was performed in MINITAB 17 to
derive relationship among the dependent (response) and inde-
pendent (predictor) variables with an acceptable statistical signif-
icance. In other words, regression expression correlates the
response variables (Yi) with predictor variables (Xi) and its general
form is given by following equation [31]:
Yi ¼ c1 X1 þ c2 X2 þ c3 X3 þ c4 X1 X2 þ c5 X1 X3 þ c6 X2 X3
þ c7 X1 X2 X3 þ b þ ε
Here, Xn denotes the predictor variables which can be in an
individual (linear regression) or product (non-linear) form. To
elaborate, consider our particular case where we have three
random variables (X1, X2, and X3). The first three terms of Equation
(7) will comprise of the single random variable as a predictor. The
fourth term will be the product of X1 and X2, demonstrating the
interaction effect of two random variables on response and so on.
Hence the total number of predictor variables will be equal to 7 in
this case (3 individual terms, 4 combined terms). Last two terms in
Equation (7) are regression constant and error term respectively.
Regression model summary mirrors the extent of accurate
representation of the sample data through regression equation.
Important parameters to consider in this category are coefficient of
determination (R2), p-value and Variance Inflation Factor. R2 value
determines the extent of variation of response variables, as pre-
dicted by the regression equation, with sample data. P-value and
variation inflation factor in context with the current study are
described in the next paragraphs.
P-value in a statistical analysis refers to the probability of finding
data points in the regression equation which prove null hypothesis
to be true [32]. Null hypothesis refers to regression result being
statistically insignificant, or in simple words, predictors and
response variables being unrelated. It is important to note that p
values smaller than 0.05 have a significant effect on the predictors.
For the correlation to be statistically significant, the p-value should
be less than confidence interval, which is assumed to be 5% for this
study [31].
Variance Inflation Factor (VIF) refers to the extent of correlation
between different predictor variables (also known as multi-
collinearity) in the regression analysis. High values of VIF repre-
sents that the predictor variables are strongly correlated with each
other and thus contribute largely towards the overall variance of
the regression expression. Several studies in the literature [33]
recommend values of VIF lower than 10 to avoid unnecessary
instability of regression coefficients.
Table 2 summarizes the statistical results of 6 and 4 terms
regression analysis of the CFD results for scenario B, whereas the
regression results for the case of individual variation (rupture
location only) are shown in Table 3. It is important to mention here
that the 7th term in equation (7) (c7X1X2X3) is not included since
initial computations resulted in a very high VIF. Results in Table 2
for 6 term regression analysis show that predictors X3 (for Q duct
outlet only), X1*X3 and X2*X3 have a p-value higher than the sig-
nificance level of 0.05. This means that these predictors can be
safely neglected in the regression equation. It is important to note
(7)
that for the sake of brevity, only two response variables, namely fan
power (Y1) and volume flow rate delivered at working face (Y4) are
shown in Table 2. The VIF for different predictors is all less than 10,
thus meeting the regression coefficient stability criteria. Coefficient
of determination (R2) values for both of the response variables are
above the acceptable limit of 85% set for this study.
Based on statistical analysis of 6 term regression equation, the
analysis was re-run after omitting the redundant predictor terms
with high p-values. The statistical summary for this run is also
presented in Table 2. A slight decrease in VIF values is observed
referring to the improved stability of regression coefficients.
However, this decrease comes at a cost. The R2 values have
decreased, albeit slightly, for 4-term regression equation. None-
theless, the values remain above the recommended limit of 85%. An
important conclusion to draw from the statistical results presented
in Table 2 is that rupture orientation has a negligible effect on all
response variables as demonstrated by high p-values. Hence it is
safe to posit that the consideration of the rupture orientation is
superfluous in this context and can be safely ignored in the
regression equations along with any further output data analysis.
For scenario A (varying rupture location only) the regression
expressions for all the output variables are shown in Table 3. The
regression expressions are relatively simpler since the predictor
variable reduces to one for this specific scenario.

Table 2
Statistical summary for 6 terms and 4 terms regression equations- Varying all parameters.

Response variables 6 terms 4 terms

2
Predictors Coefficients P-Value VIF R Coefficients P-Value VIF R2

Fan power (W) Constant 5365.4 0 94.93% 5347.2 0 94.71%

X1 5.117 0 5.9 5.212 0 5.8
X2 351.9 0 5.91 333.5 0 5.64
X3 0.628 0.042 7.82 0.071 0.525 1.02
X1*X2 1.582 0 9.91 1.488 0 9.72
X1*X3 0.00151 0.374 3.06 e e e
X2*X3 0.359 0.063 5.43 e e e

Q duct outlet Constant 11.489 0 87.18% 11.461 0 86.83%

X1 0.00556 0 5.9 0.00555 0 5.8
X2 1.388 0 5.91 1.347 0 5.64
X3 0.000346 0.717 7.82 0.000031 0.929 1.02
X1*X2 0.0053 0 9.91 0.00515 0 9.72
X1*X3 0.000005 0.32 3.06 e e e
X2*X3 0.000682 0.255 5.43 e e e
206 S. Akhtar et al. / Building and Environment 121 (2017) 200e214

Table 3 on the response variables Yi.

Statistical summary for regression analysis for X1- Varying rupture location. For scenario B, the spearman's rank correlation coefficient [34]
Response variables Predictors Coefficients P-Value R-squared is implemented in Minitab 17 to determine the relationship be-
Fan power (W) Constant 3327 e 96.53%
tween the predictor and response variables. The value of correla-
X1 14.833 0 tion coefficient varies from 1 (strong inverse dependency) to 1
X1*X1 0.0316 0 (strong increasing dependency) with the value of 0 referring to no
Leakage Percentage Constant 0.32787 e 94.33% dependency. It is important to note here that scripting capability of
X1 0.001362 0
Minitab 17 was utilized to automate the multiple regression anal-
Total Pressure jump (Pa) Constant 275.16 e 94.20%
X1 0.5631 0 ysis as well as rank coefficient evaluation of the data.
Q duct outlet (m3/s) Constant 9.9011 e 88.73% Probability distribution bar plots are developed from the sample
X1 0.01145 0 output. The statistical fits were drawn using the Scipy [35] and
Q fan inlet (m3/s) Constant 14.5852 e 95.99%
Matplotlib [36] modules as implemented in Python 3.5.1.
X1 0.008612 0

3. Model validation and verification

2.2.2. Sample output analysis
Once the regression equations for response variables are final-
ized, all the response variables values (fan operating point and
power along with duct discharge flow rate and leakage) are
computed for 4 different sample sizes between 100 and 10,000. The
reason for increasing the sample size is to get enough data points so
that the Cumulative Distribution Function comes out to be smooth.
The output variable selected for the sample sensitivity analysis
is volumetric flow rate at the discharge end of the auxiliary duct
(Y4). Fig. 4 shows the cumulative distribution function (CDF) of Y4
for different sample sizes. It is evident that the variation CDF de-
creases with the increase in sample size. Hence a sample size of
10,000 is selected for output analysis.
The output analysis for scenario A is carried out using the co-
efficient of variation as a prime indicator. The expression for the
coefficient of variation is given by:
sy
cv;i ¼
Y
Y is the statistical mean of the sample whereas sy is the standard
deviation of the sample output which is given by:
!1=2
1 XN
sY ¼ ðY  YÞ2
N i¼1 i
Equation (9) quantifies the effect of each input random variable
Fig. 4. Cumulative Distribution Function of flow rate through discharge end for
different sample sizes.
Many literature studies have validated the numerical model for
the velocity field inside the tunnel against in-situ experimental
data. A previous study by the corresponding author [20] validated
the numerical model with the experimental results of Parra et al.
[5]. Results show that the k-epsilon turbulence model yields an R2
value of 0.96 with the experimental data thus highlighting its
suitability for modeling the velocity flow field in an underground
tunnel. CFD validation conducted by other studies [24,25,37] reaf-
firm that k-epsilon suitably mirrors the 3-dimensional turbulent
velocity field in an underground drift.
It is important to mention here that fan boundary condition is
verified for the base case configuration first. Base case configuration
contains no ruptures what so ever in the auxiliary duct which
simplify the simulation for the purpose of fan boundary condition
verification.
In order to verify the validity of the fan boundary condition as
implemented in commercial CFD code, following two checks have
(8) been implemented.
1) For the base case, the operating point of a fan as calculated by
ANSYS FLUENT fan boundary condition is compared with the
intersection point of fan equation and system characteristic
curve.
2) The implementation is then extended to the ruptured duct, and
(9)
the behavior of the operating point is qualitatively compared
with that of the rupture-free duct.
The fan operating point for the base case, illustrated in Fig. 5 as
OP 1, corresponds to fan flow rate of 14.88 m3/s and a total pressure
of 385 Pa. The operating point as computed by the fan boundary
condition gives a flow rate of 14.15 m3/s and a pressure of 373 Pa.
Both flow rate and pressure agree within 5% value of each other.
This is reasonable agreement keeping in mind the statistical errors
that may have incurred due to the estimation of fan curve and
system characteristic equations. It is noteworthy that the theoret-
ical fan operating point e intersection between fan characteristic
curve (FCC) and system characteristic curve (SCC) e was derived
from experimental data. The FCC provided by fan manufacturer was
experimentally tested and measured. Similarly, the SCC was ob-
tained from Eq. (4) for which the k-factor used in the calculation k-
factor from Ref. [4] was experimentally measured from under-
ground drift. During the experiment, the airflow rate and pressure
in the underground drift were measured in order to get k-factor.
The SCC from CFD result (line) are in very good agreement with
theoretical SCC (x symbol).
Fig. 5 shows the difference in the operating point for fans
installed in leak-free versus leaky ducts. OP 1 corresponds to the
operating point (total pressure and volumetric flow rate through
the fan) for a leak free duct. Total pressure here is the sum of static
and dynamic pressure in the auxiliary duct. The solid black curve
 S. Akhtar et al. / Building and Environment 121 (2017) 200e214
Fig. 5. Characteristic curves and fan operating points for different duct topologies.
passing through OP 1 is the characteristic curve for the base case.
OP 2 and 3 show the fan operating points for a duct ruptured at the
bend with an average width (0.335 m). OP 2 is based on the flow
rate through the fan in a ruptured duct whereas OP 3 refers to the
operating point based on flow rate at the discharge end of the duct.
It is worthwhile to mention here that the fan characteristic curve
(blue line in the figure) is kept constant while comparing the
operating points for different ducts.
The difference between OP 1 and OP 2 is intuitive to predict.
Since the fan curve is inputted into the CFD code, the operating
point would follow the fan curve regardless of the whether the duct
has leakage or not. The position of OP 1 as compared to OP 2 in Fig. 5
shows that for leaky duct the flow rate through fan increases which
leads to a subtle drop in the total pressure and consequently the fan
power. This observation, though insightful, is of little practical
consequence. A more practical approach shall attempt a compari-
son of operating point based on flow rates at the discharge end of
the auxiliary duct. This is depicted by the difference between OP 1
and OP 3 in Fig. 5. OP 3 is defined as a virtual operating point for the
leaky duct based on its discharge flow rate. The dotted curve
passing through OP 3 is the characteristic curve of the leaky system
based on discharge flow rate. The purpose of this is to gain come
insights with regards to change in flow resistance with leakage in a
duct. A comparison of OP 1 and OP 3 reveals a marked difference
between the characteristic curves based on flow rate at the
discharge ends for ruptured and non-leaky ducts. The resistance of
the flow through leaky duct comes out to be significantly greater as
compared with the leak-free duct. A direct corollary of this behavior
is a drastic reduction (almost 37% in this case) in the delivered air at
the working face for the ruptured duct. In order to offset this
reduction, the flow rate needs to be increased which ultimately
entails using a fan with higher pressure, volumetric flow rate, and
consequently higher power rating.
The behavior of leak-free versus leaky ducts is consistent with
observations made by previous studies [9]. This warrants the model
to be extensively applied to a wide gamut of ruptured duct
207
scenarios. The following section presents some of the key findings
of the statistical and CFD analysis.
4. Results and discussion
This section elucidates on the results from a combined MCS and
regression analysis of CFD model for ruptured ducts. It also sum-
marizes the sensitivity of different rupture characteristics on fan
operating point, the flow rate at working face and fan power as
characterized by a coefficient of variation and probability distri-
butions for both scenarios (A and B) separately. For scenario A,
there are further two cases which are examined separately. The first
case comprises of varying rupture location along the auxiliary duct
of length 68.58 m while keeping the rupture size constant at
0.3048 m. The rupture orientation was also assumed to be fixed at
the bottom of the duct for this case. The second case comprises of
examining the impact of rupture size while keeping the location
and orientation constant.
4.1. Varying rupture location
Fig. 6(aed) visually represents the scatter of all the output var-
iables for the case of varying rupture location under scenario A. A
quick look on the visuals reveals that power, total pressure and flow
rate at discharge end generally rise with the increasing distance of
rupture location along the length of the duct from the duct inlet.
The trend for other output variables namely fan flow rate and
percentage leakage are the opposite.
In order to find out the physical reasoning behind the afore-
mentioned trends, a good starting point will be to examine the
percentage leakage trend with rupture position as illustrated in
Fig. 6 (c). The plot shows that for ruptures close to the fan inlet, the
leakage levels are low at first followed by an abrupt rise to the
maximum value. After the abrupt increase, the leakage percentage
starts decreasing uniformly, baring the region of the pipe bend. This
peculiar behavior of the percentage leakage can be ascribed to the
208 S. Akhtar et al. / Building and Environment 121 (2017) 200e214
Fig. 6. Scatter plots of (a) Fan and duct discharge flow rates, (b) Total fan pressure, (c) Percentage leakage and (d) Fan power consumption, for the case of varying rupture location
under scenario A.
mechanics of fluid flow inside the circular duct. Very near to the fan
inlet, the flow is undeveloped, thus resulting in low velocity gra-
dients in the radial direction resulting in relatively low initial
leakage values. The leakage % peaks at about 3.35 m downstream of
the fan, indicating that the viscous forces are less effective near the
duct entrance. Further downstream, as boundary layer develops,
the viscous forces characterized by friction factor start taking effect.
This causes a reduction in the pressure along the duct which results
in reduced pressure differential between fluid in the duct and
rupture. This further ensues in low levels of the flow rate through
rupture as compared with fan flow rate. As the rupture position
moves towards the end of the duct the fan flow rate approaches the
discharge flow rate as presented by the intersection point of two
plots in Fig. 6 a.
Near the bend, there are some anomalous data points which
exhibit significant deviation from the general trend. Fig. 6 (c) shows
that as rupture position approaches the bend, the leakage per-
centage decreases about 11% from the nominal value. Right at the
middle of the bend (at 22.86 m mark) the leakage percentage
suddenly spikes to around 27%. This contrasting behavior can be
attributed to the perturbations caused by velocity field due to the
presence of the bend. At the middle of the bend (45
), the flow is
forced to change directions thus resulting in an enhanced collision
with the duct wall and resulting in minor flow recirculation. This
altered velocity field behavior results in enhanced air leakage
through the auxiliary duct.
The reduction in % leakage with rupture position has several
implications on other output parameters. As illustrated in Fig. 6 a,
the fan flow rate follows exactly the same trend as that of per-
centage leakage. Discharge flow rate contrarily follows an exactly
opposite trend. This decrease in fan operating flow rate results in a
higher total operating pressure of the fan (Fig. 6 (b)) thus resulting
in higher power consumption for the fan (Fig. 6 (d)) for increasing
rupture position. From these observations, it can be concluded that
the duct bend and portion of duct closer to the fan are critical areas
in terms of discharge flow rate of the system. However, as far as the
fan power and operating pressure are concerned, the rupture on
duct length after the bend is an important consideration.
4.2. Varying rupture size
The discussion thus far has established that rupture on the bend
of the duct and near the entrance region is quite critical in terms of
the performance parameters of the auxiliary ventilation system.
Fig. 7(aed) presents the impact of varying rupture size on the
output variables while keeping the location of the rupture in the
middle of the bend. The trends shown by these figures are quite
intuitive. However, they do furnish some interesting quantitative
insights when compared with data in previous Fig. 6.
Fig. 7 (c) shows the variation of % leakage with rupture size. The
data shows that percentage leakage varies almost linearly with the
rupture size. Furthermore, the absolute values for the % leakage are
approximately two-folds higher as compared with the data for
variable rupture position. This accentuates the significant depen-
dence of rupture size on percentage leakage. On the other hand, a
trend similar in magnitude but opposite in effect is observed for
discharge flow rate. Fig. 7 (a) shows that the discharge flow rate
varies sharply with rupture size as opposed to fan flow rate.
The variation of fan flow rate with rupture size, on the other
hand, is relatively insignificant. This is highlighted in Fig. 7 (a)
 S. Akhtar et al. / Building and Environment 121 (2017) 200e214
Fig. 7. Scatter plots of (a) Fan and duct discharge flow rates, (b) Total fan pressure, (c) Percentage leakage and (d) Fan power consumption, for the case of varying rupture size under
scenario A.
where the slope of fan flow rate trend is significantly lower than the
slope for discharge flow rate. Furthermore, the fan flow rate is
observed to be relatively insensitive to the rupture sizes in excess of
0.3048 m This effect is also evident from fan operating pressure and
power as shown in Fig. 7 (b and d).
When compared with the absolute discharge flow rates for the
variable rupture location case, the flow rate for the variable rupture
size comes out to be about 40% less in magnitude for the worst case.
Furthermore, the relative change in fan power for the extreme
levels of rupture size comes out to be around 19% whereas for
variable rupture position the relative change is approximately 47%.
A similar trend is observed for fan operating pressure as well.
Hence it can be concluded that rupture size impacts the discharge
flow rate more adversely as compared to rupture location. To
conclude, it can be stated that rupture size has relatively reduced
the impact on fan operating point and power consumption as
opposed to rupture position. However, this only holds true for the
average conditions i.e. for the cases where rupture location is in the
middle of the bend (for different rupture sizes) and rupture size
correspond to average leakage percentage (for different rupture
locations).
The above-mentioned trends while useful, does not truly pro-
vide the definitive means to cross compare the relative impact of
rupture location and size on the variability of output parameters.
For this particular reason, the concept of coefficient of variance is
used to get insight with regards to the relative spread of output
parameters over their mean values. Fig. 8 (a) and (b) show the
coefficient of variation for scenario A. Results show that output
variables show a wider spread relative to their mean values for
rupture location as compared to rupture size. Furthermore, the bar
209
plot also reaffirms the earlier deduction that discharge flow rate is
more sensitive to rupture size as opposed to rupture location.
So far the trends that have been observed are based on the
assumption that only one rupture characteristics vary while others
remain constant. A more realistic approach towards the analysis of
the ruptured ducts shall incorporate the cumulative effect of the
rupture parameters. This can be achieved by generating the cases
for CFD simulation with MCS while varying all parameters simul-
taneously. The preceding sub-section presents the results for the
aforementioned scenario.
4.3. Varying all parameters simultaneously
Fig. 9(aef) demonstrate the combined impact of all random
input variables (Xi) on all of the output variables for CFD simulation
in terms of their probability distribution function (PDF). Several
existing distribution functions are tried to fit the data including
Weibull minimum, Weibull maximum, Lognormal, Normal, Error,
Chi-square, Maxwell, and Logistic. The Akaike Information Criteria
(AIC) [38] is used as a measure of goodness of fit to cross-compare
results of different distributions. Results show that for flow rate
through the fan, rupture, and leakage percentage (Fig. 9 (b, c and e))
the Weibull minimum distribution has the best fit. The normal
distribution yields acceptable fit for power and pressure (Fig. 9 (a
and f)), whereas the flow rate at working face is accurately repre-
sented by Weibull maximum distribution.
Fig. 9 (a) shows the PDF of fan power consumption alongside
with average and standard deviation of the sample output. The
average power consumption for the fan comes out to be around
5.7 kW, whereas the least and highest fan consumptions were
210 S. Akhtar et al. / Building and Environment 121 (2017) 200e214

Fig. 8. Coefficient of variance bar plots of different output parameters for scenario A.

Fig. 9. Probability distribution functions of (a) Fan Power, (b) Fan flow rate, (c) Rupture flow rate, (d) Flow rate through discharge end, (e) Leakage percentage and (f) Total fan
pressure, for the case of varying all rupture parameters simultaneously (Scenario B).
 S. Akhtar et al. / Building and Environment 121 (2017) 200e214 211

recorded at 4.77 kW and 6.495 kW respectively. Important asser-
tions can be made when these values are compared with the base
case simulation (without rupture). For the base case, the power
consumption comes out to be 5.33 kW, which is higher than 24% of
the cases. Furthermore, for these 24% cases, the leakage percentage
through the rupture is significantly higher than the average leakage
value for the data. This is a surprising observation since the power
consumption for ruptured ducts is thought to be higher as
compared to leak-free ducts. However, careful cross-comparison of
output variables show that 24% of the leakage scenarios result in
shifting of the fan operating point such that the fan pressure yields
a lower magnitude. This ultimately translates into lower flow rate
at the discharge end of the duct leading towards working face.
Thus, the price of lower power consumption of fan for some cases of
ruptured ducts is ultimately paid by reduced ventilation air flow
rate.
Flow rate at the discharge end of the duct (Y4) is an important
consideration while designing the ducting systems. Fig. 9 (d) shows
that the sample output data for Y4 is best-fitted using the Weibull
distribution skewed towards the right (Weibull maximum). The
mean and extreme values of the output variable are shown in
Table 4. Comparing the mean value for the Y4 for different rupture
scenarios with the base case value reveals that the average flow rate
is reduced by 27% for ruptured ducts. This observation can be
rephrased probabilistically by stating that under the stated input
limits of rupture location, size and orientation there is a 50% chance
that the delivered flow rate will fall short by 27% of the base case
value. It is also observed from Table 4 that the highest possible
value for different permutations of Xi is approximately 13% less
than the base case value. Hence, in contrast to power consumption,
the flow rate at the discharge end is always less than the base case
value for all ruptured ducts scenarios.
In order to quantify the sensitivity of each input parameter on
output variables, the concept of the correlation coefficient is
employed. The matrix shown in Table 5 presents the spearman-rho
correlation coefficient and p-values for all 1-1 combinations of
input and output variables. Analysis of p-values in Table 5 reaffirms
our earlier hypothesis about the irrelevance of rupture orientation
in the context of output variables. Further analysis of the data re-
veals that rupture location holds the strongest dependence on all
output variables. The output variables are also strongly correlated
with rupture size albeit around 2e3.5 times less than the rupture
location. This underscores the significance of rupture location in
comparison to other input variables in affecting output variables.
The data in Table 5 shows that rupture location is directly
correlated with Power, discharge flow rate and total pressure
whereas its correlation with leakage percentage, fan flow rate, and
rupture flow rate is inverse. The correlation dependence trend for
rupture size, on the other hand, is the exact opposite of what is
observed for rupture location. Power, total pressure, and fan flow
rate show a stronger quantitative dependence to rupture location,
whereas percentage leakage, discharge flow rate and flow rate
through rupture exhibit higher correlation with rupture size
amongst all output variables.
An important practical conclusion that can be drawn from these
observations is that the relative importance of input variable de-
pends on the output variable under consideration. For example, if
the system is to be designed based on the flow rate through
discharge duct than both input variables should be given near equal
importance during the design stage since their correlation co-
efficients are close to each other. However, in the case of fan power
being the output variable under consideration, the rupture location
will be more critical towards the output and thus should be treated
with more care.
4.4. Effect of leakage on fan operating point
This section presents a discussion on the scatter of fan operating
points for all the cases under scenario A and scenario B as computed
by CFD simulations. Fig. 10 (a, b) shows the fan operating points for
the cases of (a) varying rupture position and (b) varying rupture
sizes for scenario A on a pressure-flow rate of system characteristic
curve.
In fluid flow, the resistance is defined as the ratio between
pressure drop and the square of volumetric flow rate. This resis-
tance, as quantified by Equation (1), is dependent on fluid and flow
properties such as density, viscosity, and flow rate along with
geometrical properties of the duct such as the area of cross-section,
wetted parameter etc. Thus the total resistance of the ventilation
system with leakage will be given by Ref. [4]:

Table 4
Mean and extreme values of output variables for scenario B versus leak free duct.

Variable Average value Lowest value Highest value Base case value

Power 5721.2 4489.8 6526.1 5330

Q fan 13.6 12.6 15.3 14.146
Q Rupture 2.48 0.082 7.29 e
Q Discharge 11.16 7.99 12.53 14.146
Leakage Percentage 17.77 1.34 50 e
Total Pressure 419.7 328.3 479.9 373

Table 5
Correlation coefficients matrix of input variables versus output variables.

Input variables (Xi)

Rupture location (X1) Rupture Size (X2) Rupture Orientation (X3)

Output Variables Correlation Coefficient P-Value Correlation Coefficient P-Value Correlation Coefficient P-Value

Power 0.96 0 0.293 0.003 0.029 0.777

Q fan 0.959 0 0.296 0.003 0.03 0.765
Q Rupture 0.894 0 0.422 0 0.033 0.742
Q Discharge 0.818 0 0.507 0 0.049 0.629
Leakage Percentage 0.885 0 0.434 0 0.035 0.729
Total Pressure 0.961 0 0.292 0.003 0.03 0.765
212 S. Akhtar et al. / Building and Environment 121 (2017) 200e214

Fig. 10. Fan operating points for the cases of (a) varying rupture position and (b) varying rupture sizes for scenario A. Red markers depict plots of fan flow rate and total pressure,
whereas the plots of duct discharge flow rate versus total fan pressure are shown in yellow markers. The dotted line shows characteristic curve for the leak free duct. (For
interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

simultaneous variation of different rupture parameters, the likeli-

1 1 1 hood of higher flow resistance for leaky ventilation system in-
pffiffiffiffiffiffi ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi þ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
RT RLeakage Rleakfree
The auxiliary ventilation system with leakage can be modeled as
a superposition of a virtual duct without leakage and leakage
component of the flow. This hypothetical duct, branching out from
the original duct and carrying all the leaked flow, is assumed in
order to assign a physical meaning to the RLeakage term in equation
(10).
From equation (10) it can be deduced that the leakage causes the
overall resistance of the ventilation system to reduce assuming that
the leak free duct's flow resistance remains constant. However,
since the flow resistance of the virtual duct is dependent on friction
factor, which in turn is a function of Reynolds number, the
assumption stated afore is tentative at best. This assertion is further
consolidated by the trend observed in Fig. 10. The figure illustrates
the spread of fan operating point for the ruptured duct in com-
parison with the characteristic curve for the leak free duct. The
relative magnitude of flow resistance can be gaged from the graph
by comparing the relative positions of the operating point. Ac-
cording to equation (1), the scenarios with higher resistances will
lie above the black dotted line (characteristic curve of leak free
resistance) in Fig. 10. Correspondingly, the scenarios with lower
resistances would lie below the dotted black line. From the plots (a)
and (b) it is evident that the duct resistance both can be greater or
lower than the leak free duct case depending on the complex
interaction of rupture parameters with flow characteristics. The
data points based on flow rate at the discharge end, however, al-
ways exhibit greater resistance for any combination of rupture
variables.
A similar trend is observed in Fig. 11 which depicts the fan
operating points for scenario B. In this particular case, however,
most of the operating points lie above the characteristic curve for
leak free duct as opposed to scenario A. This shows that for
(10)
creases. Thus care must be taken while using the concept of flow
resistance in designing auxiliary ventilation systems.
5. Conclusion
The effect of rupture position, size, and orientation has been
studied, both separately and combined, on fan operating point,
power consumption and discharge flow rate. MCS in conjunction
with linear regression and CFD is used to encompass both the
random nature of rupture parameters and complex 3-Dimensional
turbulent fluid flow, into analyzing the performance of ventilation
system. Statistical and CFD results showed that the output pa-
rameters are a weak function of rupture orientation, possibly due to
the symmetry of the duct. However, rupture position and size
strongly influence the fan operating point along with leakage per-
centage and duct flow rate towards working face.
To summarize, the results highlight the qualitative and quanti-
tative differences in the dependency of output variables (fan
operating point and power along with discharge and leakage flow
rates) on rupture parameters (rupture size, location, and orienta-
tion) for both leakage scenarios (separate and simultaneous vari-
ation). For an individual variation of the leakage parameters, fan
operating point and power consumption was found out to be
extremely sensitive to the location of the rupture on the duct. On
the other hand, rupture size showed a dominant impact on per-
centage leakage and flow rate at the discharge end. Leakage
magnitude as high as 60% and a decrease in discharge flow rate of
around 47% were recorded across the extremities of rupture size.
The data also shows that rupture at the center of the bend is critical
and leads to higher flow rate loss and fan power consumption.
For the simultaneous variation of parameters, the probability
density functions of the output variables show a chance of 76% for
fan power consumption being greater than that for the leak free
 S. Akhtar et al. / Building and Environment 121 (2017) 200e214 213

Fig. 11. Fan operating points for the case of simultaneous variation of all rupture parameters.

duct. However, the discharge end flow rate will always be less than u Velocity field of the air throughout the computational
the leak free duct for any case of the ruptured duct. The probability domain
distribution for the discharge flow rate is skewed to the right which m Dynamic viscosity of air
demonstrates a higher likelihood for the flow rate to lie towards the s Deviatoric stress tensor
higher end of the range. x Cartesian coordinate
The results also elucidate that for a given fan, the overall resis- Y Response variable
tance of the leaky duct can be both greater and lower than that of X Input variable
the leak free duct. The reasoning for this lies in challenging the cv Coefficient of variation
assumption which presumes that in a parallel resistance network of sy Standard Deviation
the leaky duct, the flow resistance of virtual leak free duct is con- Y Statistical mean of the sample
stant. However, since the rupture characteristics influence the fan N Sample size
operating point which in turn impact the flow behavior of the duct, R Leakage resistance
the assumption of constant duct resistance is tenuous. Thus care
must be taken while basing the ventilation system design and fan Subscripts
selection based on the traditional resistance concept for leaky i, j, k Free index referring to Cartesian coordinate
ducts. i, n Number of response/input variables
T Total
Leakage Duct with leakage
Acknowledgment
Leak-free Leak free duct
turb Turbulent
The authors thank the Natural Sciences and Engineering
Research Council of Canada (NSERC) for supporting this research
Abbreviations
(241001 NSERC RGPIN-2015-03945). The simulations were con-
CDF Cumulative Distribution Function
ducted at McGill HPC facility under Calcul-Quebec and Compute-
CFD Computational Fluid Dynamics
Canada with contribution from CFI-JELF. The first author thanks
CFM Cubic Foot per Minute
to McGill Engineering Doctoral Award (MEDA) for supporting this
GB Giga Byte
research.
MCS Monte Carlo Simulations
PDF Probability Distribution Function
Nomenclature RAM Random Access Memory
RPM Revolutions Per Minute
p pressure SIMPLE Semi-Implicit Method for Pressure Linked Equations
k Darcy Weisbach friction factor VIF Variance Inflation Factor
L Length of the tunnel
per Perimeter of the tunnel's cross-section References
A Cross-section area of the tunnel
Q Air flow rate [1] T. Kalamees, Ü. Alev, M. Parnalaas, Air leakage levels in timber frame building
envelope joints, Build. Environ. 116 (2017) 121e129.
v Velocity through fan [2] G. Guyot, J. Ferlay, E. Gonze, M. Woloszyn, P. Planet, T. Bello, Multizone air
r Density of air leakage measurements and interactions with ventilation flows in low-energy
214 S. Akhtar et al. / Building and Environment 121 (2017) 200e214
homes, Build. Environ. 107 (2016) 52e63.
[3] S. Krishnamoorthy, M. Modera, Impacts of duct leakage on central outdoor-air
conditioning for commercial-building VAV systems, Energy Build. 119 (2016)
340e351.
[4] M.J. Mc Pherson, Subsurface Ventilation and Environmental Engineering,
Springer Science & Business Media, 2012.

ndez, Numerical and experimental
[5] M.T. Parra, J.M. Villafruela, F. Castro, C. Me
analysis of different ventilation systems in deep mines, Build. Environ. 41
(2006) 87e93.
[6] A. Widiatmojo, K. Sasaki, N.P. Widodo, Y. Sugai, A.Y. Sahzabi, R. Nguele, Pre-
dicting gas dispersion in large scale underground ventilation: a particle
tracking approach, Build. Environ. 95 (2016) 171e181.
[7] M. Bascompta Massane
s, L. Sanmiquel Pera, J. Oliva Moncunill, Ventilation
management system for underground environments, Tunn. Undergr. Space
Technol. 50 (2015) 516e522.
[8] A.P. Sasmito, E. Birgersson, H.C. Ly, A.S. Mujumdar, Some approaches to
improve ventilation system in underground coal mines environment e a
computational fluid dynamic study, Tunn. Undergr. Space Technol. 34 (2013)
82e95.
[9] G. Auld, An estimation of fan performance for leaky ventilation ducts, Tunn.
Undergr. Space Technol. 19 (2004) 539e549.
[10] S.J. Schatzel, R.B. Krog, A. Mazzella, C. Hollerich, E. Rubinstein, A study of
leakage rates through mine seals in underground coal mines, Int. J. Min.
Reclam. Environ. 30 (2016) 165e179.
[11] F. Colella, G. Rein, R. Borchiellini, R. Carvel, J.L. Torero, V. Verda, Calculation
and design of tunnel ventilation systems using a two-scale modelling
approach, Build. Environ. 44 (2009) 2357e2367.
[12] E. Topuz, S. Bhamidipati, M. Bartkoski, Improvement of ventilation system
efficiency through the analysis of air leakage, Proceedings of the first mine
ventilation symposium, 1982.
[13] T. Kalamees, Air tightness and air leakages of new lightweight single-family
detached houses in Estonia, Build. Environ. 42 (2007) 2369e2377.
[14] C. Belleudy, A. Kayello, M. Woloszyn, H. Ge, Experimental and numerical in-
vestigations of the effects of air leakage on temperature and moisture fields in
porous insulation, Part 2, Build. Environ. 94 (2015) 457e466.
[15] J. Jokisalo, J. Kurnitski, M. Korpi, T. Kalamees, J. Vinha, Building leakage,
infiltration, and energy performance analyses for Finnish detached houses,
Build. Environ. 44 (2009) 377e387.
[16] J.-K. Lee, Y.J. Wang, Effects of stopping leakage on coal mine ventilation sys-
tems, Int. J. Min. Eng. 2 (1984) 253e259.
[17] J. Wolski, J. Barry, Analysis of Multi-Fan Ventilation Duct Line: Resistance,
Leakage, Fan Performance, 6th International Mine Ventilation Congress, 1997.
[18] M. Onder, S. Sarac, E. Cevik, The influence of ventilation variables on the
volume rate of airflow delivered to the face of long drivages, Tunn. Undergr.
Space Technol. 21 (2006) 568e574.
[19] M. Onder, E. Cevik, Statistical model for the volume rate reaching the end of
ventilation duct, Tunn. Undergr. Space Technol. 23 (2008) 179e184.
[20] R. Whalley, A. Abdul-Ameer, The dynamics of cyclic ventilation systems, Appl.
Math. Model. 35 (2011) 93e108.
[21] G.K. Batchelor, An Introduction to Fluid Dynamics, Cambridge university
press, 2000.
[22] A.P. Sasmito, J.C. Kurnia, E. Birgersson, A.S. Mujumdar, Computational evalu-
ation of thermal management strategies in an underground mine, Appl.
Therm. Eng. 90 (2015) 1144e1150.
[23] J.C. Kurnia, A.P. Sasmito, A.S. Mujumdar, Simulation of a novel intermittent
ventilation system for underground mines, Tunn. Undergr. Space Technol. 42
(2014) 206e215.
[24] J.A. Toran~ o, R. Rodriguez, I. Diego, J.M. Rivas, A. Pelegry, Influence of the pile
shape on wind erosion CFD emission simulation, Appl. Math. Model. 31 (2007)
2487e2502.
[25] J.C. Kurnia, A.P. Sasmito, A.S. Mujumdar, Dust dispersion and management in
underground mining faces, Int. J. Min. Sci. Technol. 24 (2014) 39e44.
[26] J.C. Kurnia, A.P. Sasmito, A.S. Mujumdar, CFD simulation of methane disper-
sion and innovative methane management in underground mining faces,
Appl. Math. Model. 38 (2014) 3467e3484.
[27] J. Toran~ o, S. Torno, M. Mene
ndez, M. Gent, Auxiliary ventilation in mining
roadways driven with roadheaders: validated CFD modelling of dust behav-
iour, Tunn. Undergr. Space Technol. 26 (2011) 201e210.
[28] D.M. Hargreaves, I.S. Lowndes, The computational modeling of the ventilation
flows within a rapid development drivage, Tunn. Undergr. Space Technol. 22
(2007) 150e160.
[29] B.R. Munson, D.F. Young, T.H. Okiishi, Fundamentals of Fluid Mechanics, vol. 3,
1990. New York.
[30] C. Aydin, B. Ozerdem, Air leakage measurement and analysis in duct systems,
Energy Build. 38 (2006) 207e213.
[31] R.A. Johnson, D.W. Wichern, Applied Multivariate Statistical Analysis, Prentice
hall Englewood Cliffs, NJ, 1992.
[32] J.F. Hair, W.C. Black, B.J. Babin, R.E. Anderson, R.L. Tatham, Multivariate Data
Analysis, Pearson Prentice Hall Upper Saddle River, NJ, 2006.
[33] R.M. O’brien, A caution regarding rules of thumb for variance inflation factors,
Qual. Quantity 41 (2007) 673e690.
[34] J.H. Zar, Significance testing of the Spearman rank correlation coefficient,
J. Am. Stat. Assoc. 67 (1972) 578e580.
[35] E. Jones, T. Oliphant, P. Peterson, SciPy: Open Source Scientific Tools for Py-
thon, vol 73, 2001, 2015, p. 86. http://www.scipy.org.
[36] J.D. Hunter, Matplotlib: a 2D graphics environment, Comput. Sci. Eng. 9 (2007)
90e95.
[37] S. Torno, J. Toran ~ o, M. Ulecia, C. Allende, Conventional and numerical models
of blasting gas behaviour in auxiliary ventilation of mining headings, Tunn.
Undergr. Space Technol. 34 (2013) 73e81.
[38] K. Yamaoka, T. Nakagawa, T. Uno, Application of Akaike's information crite-
rion (AIC) in the evaluation of linear pharmacokinetic equations,
J. Pharmacokinet. Biopharm. 6 (1978) 165e175.