A Mixed Convection Model for Estimating the

Critical Velocity to Prevent Smoke Backlayering in

Tunnels
Michael Beyer

Stacey Agnew https://orcid.org/0009-0003-9433-5679

Conrad Stacey
Stacey Agnew https://orcid.org/0009-0005-0216-9103
Günter Brenn
Fluid Dynamics and Heat Transfer https://orcid.org/0000-0001-7576-0790

Research Article

Keywords: critical velocity, tunnel safety, tunnel smoke control, backlayering, re simulation

Posted Date: April 18th, 2024

DOI: https://doi.org/10.21203/rs.3.rs-4278205/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License

Additional Declarations: The authors declare no competing interests.

 A Mixed Convection Model for Estimating the
Critical Velocity to Prevent Smoke
Backlayering in Tunnels

Michael Beyer 1,*, Conrad Stacey 2 and Günter Brenn 3

1
Stacey Agnew Pty Ltd; [email protected]
2
Stacey Agnew Pty Ltd and Delve Stacey Agnew LLC; [email protected]
3
Graz University of Technology; [email protected]
*Correspondence: [email protected]

Abstract: A novel mathematical model for the critical ventilation velocity to pre-
vent smoke backlayering in tunnels is presented, addressing limitations of prior ap-
proaches. The basis of the model is a rigorous characterisation of the physical pro-
cesses by the characteristic quantities. Empirical parameters within the new model
are determined, to align with results from both full-size and small-scale tunnel ex-
periments. Data from numerical simulations (CFD, Computational Fluid Dynam-
ics), validated by known test data, are then used to estimate the effects of tunnel
slope and other parameters on the critical velocity. The model is seen to approxi-
mate the critical velocity well, following all trends identified by test data and CFD
parameter studies. The empirically calibrated equation permits prediction of the
critical velocity beyond the narrow range of tunnel geometries where known results
already give an answer. The resulting equation has practical application for tunnel
design.

Keywords: critical velocity; tunnel safety; tunnel smoke control;

backlayering, fire simulation

Nomenclature

Latin symbols
𝐴 Tunnel cross-sectional area (m2)
𝑎 Parameter used for non-dimensional trend lines (-)
𝑎𝑤 Empirical constant to offset the ‘plume blockage’ (-)
function. See Table 2 for pool fires, eq. (30).
𝑏 Maximum nondimensional critical velocity value (-)
for different non-dimensional trend lines
𝑐𝑝 Specific isobaric heat capacity based on average (J/kg K)
temperature between approaching air and mean
temperature just downstream of the fire, according
to equation (4).
𝑑𝑏 Burner diameter (m)
̅
𝐷ℎ , 𝐻 Tunnel hydraulic diameter (m)
2

𝑓𝑎 Empirical constant used in fire length function. (-)

See Table 2 for pool fires.
𝑓𝑔 Empirical constant to bias the ‘plume blockage’ (-)
function. See Table 2 for pool fires, eq. (30).
𝑓ℎ Empirical factor relating 𝐿𝑛 to the length of that (-)
part of the fire that influences backlayering pro-
pensity. See Table 2 for pool fires.
𝑓𝑜 Empirical constant used in fire length function. (-)
See Table 2 for pool fires.
𝐹𝑖 Inertial force (N)
𝐹𝑏 Buoyancy force (N)
𝑔 Gravitational acceleration (m/s2)
𝑔𝑖 Components of the gravitational acceleration for (m/s2)
the Cartesian coordinate direction 𝑖
ℎ, 𝐻 Tunnel height at fire location (from floor to high- (m)
est point at ceiling)
𝐻𝐷𝐾 Height given different meanings in different refer- (m)
ences. Distance from the base of the fire source to
the highest point in the ceiling in Danzinger &
Kennedy [6] and in Annex D of NFPA 502 version
2023 [48]. Height of duct or tunnel at fire site in
Annex D of NFPA 502 version 2014 [44]
𝐻𝑠𝑐𝑎𝑙𝑒 Tunnel height of a scaled tunnel (m)
𝐼𝑓𝑖𝑟𝑒 Fire intensity (heat release rate per unit area) (e.g. (W/m2)
pool fire with oil fuel 2.25 MW/m2 [27])
𝑘 Thermal conductivity (W/mK)
𝐾𝐹 Function for plume width influence on critical ve- (-)
locity
𝐾𝑔 Grade factor, see Table 5 (-)
𝐾𝑔,𝐷𝐾 Grade factor as defined by [6] (-)
𝐾𝐿 Fire length function (-)
𝑘𝑖𝑛𝑘 Non-dimensional total heat release rate where kink (-)
in non-dimensional trend lines happens
𝐿 Characteristic length of the problem (m)
𝐿𝑓𝑖𝑟𝑒 Total length of a continuous fire source (m)
𝐿𝑛 Characteristic length (height) for natural convec- (m)
tion (for pool fires: tunnel height from the base of
the fire to highest point at the ceiling)
𝑚 Spreading rate of rising plume relevant for interac- (-)
tion of ambient air with plume. See Table 2 for
pool fires.
𝑝∗ Non-dimensional pressure defined as (-)
𝑝∗ = 𝑝/(𝜌𝑈 2 )
𝑄̇ Total fire heat release rate (W)
𝑄̇𝑠𝑐𝑎𝑙𝑒 Total fire heat release rate in a scaled tunnel (W)
𝑄̇ ∗ Nondimensional heat release rate (-)
𝑠 Tunnel slope (ratio of height difference to length) (-)
𝑠𝑡 Transverse tunnel slope (ratio of height difference (-)
to transverse length)
𝑇 Temperature (K)
∆𝑇 Temperature difference relevant for the buoyancy (K)
force or temperature increase of a heat capaci-
tance flow rate caused by a heat release rate 𝑄̇
𝑡∗ Non-dimensional time (-)
𝑇𝑎 Temperature of upstream air (K)
𝑇𝑝 , 𝑇𝑓 Temperature of the hot gases (K)
 3

𝑈 Free stream velocity upstream of the fire or char- (m/s)

acteristic velocity
𝑈𝑐 Critical velocity (m/s)
𝑈𝑐∗ Nondimensional critical velocity (-)
𝑈𝑐_𝑓𝑙𝑎𝑡 Critical velocity in a flat tunnel (m/s)
𝑈𝑐𝑜𝑛𝑓 Confinement velocity (m/s)
𝑈𝑠𝑐𝑎𝑙𝑒 Upstream air velocity in a scaled tunnel (m/s)
𝑢𝑖∗ , 𝑢𝑗∗ Non-dimensional velocity in Cartesian coordinate (-)
direction 𝑖 and j respectively
𝑊𝑓𝑖𝑟𝑒 Maximum width of the fire source (e.g. fire pans (m)
of Memorial Tunnel tests were 3.66 m wide)
𝑥𝑖∗ , 𝑥𝑗∗ Non-dimensional distance in Cartesian coordinate (-)
direction 𝑖 and j respectively

Non-dimensional numbers
𝐸𝑐 Eckert number (-)
𝐹𝑟 Froude number (-)
𝐹𝑟𝑐 Critical Froude number (-)
𝐹𝑟𝑐,𝐷𝐾 Critical Froude number as defined by [6], 𝐹𝑟𝑐,𝐷𝐾 = (-)
4.5
𝐺𝑟 Grashof number in vertical coordinate direction (3 (-)
or z)
̃
𝐺𝑟 Modified Grashof number (-)
𝐺𝑟𝑖 Grashof number in Cartesian coordinate direction 𝑖 (-)
𝑀 Mach Number (-)
𝑃𝑟 Prandtl number (-)
𝑅𝑒 Reynolds number (-)
𝑅𝑖 Richardson number (-)

Greek symbols
𝛽 Isobaric thermal expansion coefficient (-)

𝛾𝐿 Empirical constant of fire length function. See Ta- (-)

ble 2 for pool fires.
𝜀 Heat reduction due to imperfect combustion and (-)
thermal radiation
𝜂 Reduction of the convective heat release rate due (-)
to water mist systems or sprinklers
Θ∗ Non-dimensional temperature defined as Θ∗ = (-)
(𝑇 − 𝑇𝑎 )⁄∆𝑇
𝜇 Dynamic viscosity (Pa s)
𝜈𝑎 Kinematic viscosity of the approaching air (m2/s)
𝜈𝑝 Kinematic air viscosity at the mean gas tempera- (m2/s)
ture in the plume
𝜌𝑎 Density of upstream air (kg/m3)
Φ∗ Non-dimensional dissipation function (dissipation (-)
to heat due to viscous stresses)
4

1 Introduction

”Critical velocity” is that speed of oncoming tunnel air which just prevents up-
stream backlayering of smoke in a tunnel fire. Achieving critical velocity as a min-
imum airspeed in fire response is often a design criterion for tunnel ventilation, par-
ticularly in the US, Australia and some countries in Asia. There is generally no
practical upper limit in those jurisdictions. Other jurisdictions do not see this critical
velocity approach as the best or recommended approach [50]. European approaches
[38-43] for example encourage more moderate air speeds, acknowledging that risks
from over-ventilating are also significant. The US standard NFPA 502 historically
relied on the critical velocity philosophy, but recently moved away from preventing
backlayering to controlling backlayering [48], allowing some extent of smoke back-
layering. However, even if absolute prevention of backlayering from a tunnel fire
should not always be the first choice during the self-rescue phase, it can still be a
useful design or benchmark criterion. Therefore, there is motivation to have a reli-
able tool for estimating critical velocity.
In the published literature, there have been three approaches to get to the answer:
(i) conducting full-scale fire tests (expensive and hence rare), (ii) using a Computa-
tional Fluid Dynamics (CFD) model validated against full-scale tests, or (iii) calcu-
lating the critical velocity using accepted semi-empirical equations.
Over the last decades, many full-scale fire tests were carried out, but few of them
were designed to accurately determine critical velocity in tunnels for large fires.
The best-known full-scale data for critical velocity with a wide and relevant range
of heat release rates (HRRs) are the Memorial Tunnel fire tests [27-29]. These tests
are well documented and thus well suited for validation. Several authors have com-
pared different CFD models and approaches with the Memorial Tunnel full-scale
tests previously (see Karki et. al. [35], Rhodes [55] and Galdo Vega [56]). For the
purpose of the present study, a new CFD model validated by data from the Memo-
rial Tunnel fire tests was developed and recently presented by Beyer and Stacey
[51]. Conducting CFD simulations can be time consuming and requires knowledge
about the software and the choice of physical models and boundary conditions but,
when validated appropriately, is able to provide reasonably accurate answers for a
specific tunnel project with tunnel geometry and fire scenario different from the
known full-scale data. It also allows a “confinement velocity”, i.e., the air speed
allowing for a designated length of hot smoke backlayering under the tunnel ceiling,
to be determined.
The simplest and fastest way for estimating a critical velocity is the use of a well-
validated semi-empirical equation. There are many semi-empirical approaches
based on dimensional or non-dimensional models [1-21]. The results and answers
of those approaches can be very different. However, only a few of them have been
adopted in versions of the US standard NFPA 502, which confers contractual rele-
vance for real road tunnel projects. Therefore, more focus is laid on these calcula-
tion methods for critical velocity. Some of the references noted above [1-21] are
summarised and discussed in Section 2. For a more detailed explanation of the
different methods, reference is made to other work [22-26].
For many years, Annex D of NFPA 502 [44] proposed an equation introduced
by Kennedy [6]. Even though the equation proposed by Kennedy was mainly based
on the Memorial Tunnel fire test results, the equations in the Annex D of NFPA 502
were first amended [45] and then replaced [46] to better represent small-scale test
results published by Li et al. [18-21].
 5

The work by Li et al. [18-21] overpredicts the critical velocity [36, 47, 49], for
reasons established in several papers[36, 49, 52, 53]. That also explains why Ken-
nedy [6] and Li et al. [18-21] could both claim support from Memorial Tunnel data
for very different critical velocity values. The representation of the Memorial Tun-
nel data by Kennedy [6] was correct.
As discussed above, there has been conflicting and confusing guidance from
NFPA 502 for designers on estimating critical velocity of smoke backlayering in
tunnel fires. The 2023 edition of NFPA 502 [48] did not achieve the required clar-
ity. One recommendation to designers was to look at the original test data for values
appropriate to a particular tunnel (Beyer et. al. [50]). However, even with real test
data available, it is still helpful to have a physically based mathematical relation
allowing the critical velocity to be estimated for tunnel geometries or cases not well
represented by known tests.
The present work reports such a mathematical formulation for critical velocity
and its validation. Section 2 reviews the current methodology. Section 3 presents
the proposed physical model, and Section 4 provides the CFD validation and a sys-
tematic parameter variation study to test the derived physical model. In Section 5,
the model is validated against real test data. As the equation in Section 3 is perhaps
too complex for design purposes, Section 6 provides a simpler, practical estimation
of the critical velocity for design purposes. Conclusions of the work are summa-
rized in Section 7.

2 Current Methodology

The present section reviews frequently discussed methods for calculating the
critical velocity. Several articles that provide a comprehensive overview are already
available [22-26] and they are not repeated here. The methods are based on char-
acteristic numbers of fluid mechanics, which are briefly reviewed. The models es-
sentially yield the required critical velocity as a dimensional result.

2.1 Froude Number

The Froude number 𝐹𝑟 is a non-dimensional number representing the ratio of the

inertial force to the gravity force;
𝑈
𝐹𝑟 = (1)
√𝑔𝐿
where 𝑔 is the gravitational acceleration, 𝑈 a characteristic velocity and 𝐿 a
characteristic length scale of the flow field. In fluid mechanics, the Froude num-
ber is significant for flows with free surface effects, like water waves in open
channel flows, and for any isothermal fluid flow, which can be significantly influ-
enced by the body force due to gravity. In the tunnel community and in fire engi-
neering, a different form and definition of the Froude number, sometimes called
the ‘critical Froude number’, can be found. The distinction is discussed in the fol-
lowing Section 2.2.
6

2.2 Dimensional Model

One of the first calculation models for the critical velocity was introduced by
Thomas [1-2]. He concluded that the upstream spread of hot smoke in tunnel fires
can be described by the ratio between the inertial force (velocity head) and the buoy-
ancy force (buoyancy head). It was assumed that, at the critical condition where
backlayering of hot smoke is just prevented, the stagnation pressure (velocity) head
will be of the same order of magnitude as the buoyancy head. The inertial force, 𝐹𝑖
and the buoyancy force, 𝐹𝑏 , as defined by Thomas [1-2] are:
𝐹𝑖 = 𝜌𝑎 𝐿2 𝑈 2 (2)

𝐹𝑏 = 𝜌𝑎 𝑔𝐿3 𝛽∆𝑇 (3)

where 𝐿 is a characteristic length scale in the problem, 𝛽 is the isobaric thermal
expansion coefficient of the gas, and ∆𝑇 is the temperature difference relevant for
the buoyancy force. The approach taken by Thomas determines ∆𝑇 in the buoy-
ancy as the temperature increase of a heat capacitance flow rate 𝑐𝑝 𝜌𝑎 𝐴𝑈 caused
by the heat release rate (HRR) 𝑄̇ , as it follows from the First Law of thermody-
namics for an open system.
𝑄̇
∆𝑇 = (4)
𝑐𝑝 𝜌𝑎 𝐴𝑈
Here, 𝐴 is the tunnel cross-sectional area and 𝑐𝑝 the specific isobaric heat capac-
ity of the gas. By expressing the temperature rise as a function of air speed, the
equation above connects the temperature increase, and hence the buoyancy force,
to inertia effects. Substituting this equation into equation (3) and forming the ratio
of inertia force to buoyancy force (𝐹𝑖 /𝐹𝑏 ) gives the Froude number as defined by
Thomas [1].
𝜌𝑎 𝑐𝑝 𝐴𝑈 3
𝐹𝑟𝑐 = (5)
𝑔𝐿𝛽𝑄̇
The number above is also named the ‘critical Froude number’, but in fluid dy-
namics, the number is a Richardson number 𝑅𝑖 (i.e., a ratio of Grashof number 𝐺𝑟
to Reynolds number squared 𝑅𝑒 2 ) rather than a Froude number as given by equation
(1). In this approach there is only one characteristic length scale, other than that
embodied in the tunnel cross-sectional area A. As a result, the method does not
distinguish between tunnel height or hydraulic diameter, or the fire plume height.
The length scale here is taken to be the tunnel height, 𝐿 = 𝐻. Thomas then con-
cluded that the critical Froude number is expected to be in the order of unity at the
flow situation where backlayering of hot smoke is just prevented, arriving at a crit-
ical velocity formula of
1⁄3
𝑔𝐻𝑄̇
𝑈𝑐 = ( ) (6)
𝜌𝑎 𝐴𝑐𝑝 𝑇𝑝
where 𝛽 has been replaced by the inverse of the temperature of the hot gas,
1/𝑇𝑝 , treating the gas as ideal.
A similar equation was proposed by Danziger and Kennedy [4, 6]. In contrast to
equation (6), the total HRR 𝑄̇ was replaced by the convective HRR, 𝑄̇ (1 − 𝜀), with
the fractional convective heat reduction 𝜀 due to radiation. A correction for the
 7

tunnel slope was also introduced. Furthermore, the temperature 𝑇𝑝 was replaced by
the bulk temperature downstream of the fire, 𝑇𝑓 . This led to the critical velocity
1⁄3
−1⁄3 𝑔𝐻𝐷𝐾 𝑄̇ (1 − 𝜀)
𝑈𝑐 = 𝐾𝑔,𝐷𝐾 𝐹𝑟𝑐,𝐷𝐾 ( ) (7)
𝜌𝑎 𝐴𝑐𝑝 𝑇𝑓
with the grade correction factor for downgrade slopes 𝑠 given as
𝐾𝑔,𝐷𝐾 = 1 + 0.0374 ∙ [𝑎𝑏𝑠(𝑠) ∙ 100]0.8 (8)
The proposed grade correction factor was derived from studies on the control of
methane layers in coal mines. It was suggested that the grade correction should
only be used for ventilating downgrade. For uphill grades, 𝐾𝑔,𝐷𝐾 is 1.0.
The temperature 𝑇𝑓 in equation (7) is the bulk temperature downstream of the
fire and is expressed as
𝑄̇ (1 − 𝜀)
𝑇𝑓 = + 𝑇𝑎 (9)
𝑐𝑝 𝜌𝑎 𝐴𝑈𝑐
where 𝑇𝑎 is the approaching ambient air temperature. For the critical Froude
number, Kennedy [6] used a conservative value of 4.5 instead of unity as Thomas
proposed. This conclusion resulted from the observations by [5].
Based on that value, Kennedy [6] validated the derived formula against the lon-
gitudinal ventilation test results of the Memorial Tunnel Fire Ventilation Test Pro-
gram (MTFVTP) [27-28] and concluded an under-prediction of the critical velocity
for low fire heat release rates (<20 MW) and an over-prediction for high heat release
rates. The proposed formula was also adopted into Annex D of NFPA 502 [44].
Attention needs to be paid on the definition of 𝐻𝐷𝐾 in the different quotations of
this equation. The proposed formula by Kennedy [6] uses for 𝐻𝐷𝐾 the distance from
the base of the fire source to the highest point in the ceiling. The 2014 NFPA 502
Annex D uses the height of duct or tunnel at fire site (𝐻 ) for 𝐻𝐷𝐾 . From a design
perspective, it makes sense to use the full tunnel height for 𝐻𝐷𝐾 as the height of a
design fire is usually not specified, and it is a more onerous case to assume that the
base of the fire is at the roadway. The 2023 version of Annex D [48] also references
the 2014 Annex D equation, but with 𝐻𝐷𝐾 inexplicably redefined as being ‘height
from the base of the fire to the tunnel ceiling at the fire site (not tunnel height 𝐻 )’
without providing guidance on a typical fire height for a design fire.

2.3 Non-dimensional Froude Model

Non-dimensional Froude Models refer to the Froude number as defined by equa-

tion (1) and not to the ‘critical Froude number’ or the Richardson number. The
former neglects density variation and the latter two include density variation.
Non-dimensional Froude models are based on a non-dimensional critical veloc-
ity 𝑈𝑐∗ and HRR 𝑄̇ ∗ , which are defined as
𝑈𝑐
𝑈𝑐∗ = (10)
√𝑔𝐻

𝑄̇
𝑄̇ ∗ = (11)
𝑐𝑝 𝜌𝑎 𝐻2 √𝑔𝐻 𝑇𝑎
8

Both numbers use the reference velocity √𝑔𝐻 for the non-dimensionalisation. In
analogy to the definition of the Froude number in equation (1), equation (10) defines
the non-dimensional velocity, and the velocity 𝑈 in equation (4) is also replaced by
√𝑔𝐻. The tunnel cross-sectional area 𝐴 is represented by 𝐻 2 , and the temperature
difference ∆𝑇 by the ambient temperature 𝑇𝑎 . Relations between the non-
dimensional critical velocity and fire heat release rate are exclusively investigated
in model experiments at a scale as small as 1:20, as demonstrated by Oka and
Atkinson [13], Wu and Bakar [15] and Li et al. [18]. In all these investigations,
Froude similarity for scaling the velocity and fire HRR was assumed as follows

𝑈 𝑈𝑠𝑐𝑎𝑙𝑒
𝐹𝑟 = = (12)
√𝑔𝐻 √𝑔𝐻𝑠𝑐𝑎𝑙𝑒
where the index 𝑠𝑐𝑎𝑙𝑒 refers to the values in the scaled tunnel model. In the
work of Oka and Atkinson [13], as well as in Li et al. [18], 𝐻 is the duct (tunnel)
height, but in Wu and Bakar [15], tunnel height 𝐻 is replaced by the hydraulic di-
ameter of the tunnel 𝐻̅ . The scaling as given by equation (12) leads to the follow-
ing relationship for velocity and heat release rate:

𝑈 𝐻
=√ (13)
𝑈𝑠𝑐𝑎𝑙𝑒 𝐻𝑠𝑐𝑎𝑙𝑒

1
𝑄̇ 𝐻 (2+ )
2
=( ) (14)
𝑄̇𝑠𝑐𝑎𝑙𝑒 𝐻𝑠𝑐𝑎𝑙𝑒
In the definition of the Froude number used by those authors, buoyancy is not
included. Consequently, this approach inherently presumes that the temperature
difference, and therefore the density deficit 𝛽∆𝑇 between the scaled and the real
tunnel, is the same. Moreover, it also assumes that the geometric proportions be-
tween area, hydraulic diameter and tunnel height are the same between the scaled
and the real tunnel. For example, if the width to height ratio of the real tunnel is not
the same as in the scaled duct, the mass flow rate, and therefore the gas temperature
after the fire, are not scaled in the right manner. The same applies for the ratio
between hydraulic diameter and tunnel height, which are relevant for the ratio of
inertia to buoyancy force.
Beyond the velocity and the HRR, the fire intensity (heat release rate per unit
area) also needs to be scaled in the right manner to avoid altering the plume density
deficit between the real and the scaled tunnel. Failure to attend to the fire intensity
would violate the assumption that allows Froude scaling to even be considered.
This also includes the need to scale the flame length and plume volume in the right
proportions.
The test setups of the different small-scale tests [13-15, 18] are very similar. Wu
and Bakar [15], as well as Li et al. [18], used a fixed burner size for all tests in the
same test tunnel. The fuel flow rate was increased to increase the fire HRR, which
automatically increased the heat release rate by unit area (fire intensity). Oka and
Atkinson [13] also investigated different burner sizes and arrangement and derived
a set of equations to account for the different results. The trendline for the critical
velocity dependence on the HRR is different for the individual tests performed, but
always has a similar characteristic as given below.
𝑈𝑐∗ = 𝑎 ∙ 𝑄̇∗1⁄3 𝑓𝑜𝑟 𝑄 ∗ ≤ 𝑘𝑖𝑛𝑘
(15)
𝑈𝑐∗ = 𝑏 𝑓𝑜𝑟 𝑄 ∗ > 𝑘𝑖𝑛𝑘
 9

The parameters 𝑎, 𝑏 and 𝑘𝑖𝑛𝑘 for the different results are listed in Table 1. As
discussed above, and detailed by Stacey and Beyer [36], it is unlikely that all essen-
tial similarities are preserved in the scaled model, and hence unlikely that the trend
lines provide relevant numbers when scaled to real tunnels.

Table 1. Parameters for non-dimensional trend lines from different scale model tests.
Authors 𝒂 𝒃 𝒌𝒊𝒏𝒌
Oka and Atkinson [13] (0.22 𝑡𝑜 0.354)1 ∙ (0.12)−1⁄3 (0.22 𝑡𝑜 0.354)1 0.12
Wu and Bakar [15] (0.4) ∙ (0.2)−1⁄3 0.4 0.2
Li et al. [18] 0.81 0.43 0.15
1Value depends on the burner size and burner arrangement, see Oka and Atkinson
[13].

Oka and Atkinson [13] and Wu and Bakar [15] used a water spray device to cool
the tunnel wall near the fire source, which may have altered the heat loss from the
smoke layer.
Any approach that does not include the density difference cannot be regarded as
a physically reasonable approach for approximating density-driven flow behaviours
with the confidence needed for design.

2.4 Practical physical problems with prior approaches

There are several observations to be made on the limitations of models previ-

ously applied.
The first is the dependence of the critical velocity on the heat release rate. That
was a problem for Kennedy et al. [6] in matching the Memorial Tunnel data [27-
30], and has been a problem since. Looking at the data relevant to road tunnel
design, such a dependence could not be identified. Above a certain fire heat release
rate, there is no further increase in critical velocity (see Figure 16).
Related to the problem of the heat release dependence is the second problem,
which is the influence of the fire intensity. Fire intensity depends on the fuel type
(heat of combustion) and the ability of the fuel to mix and burn before being advec-
ted downstream (burning rate). If the burning takes place while advecting down-
stream, the fire area has been increased and the intensity reduced. So, for a given
fire type, an increase in the fire heat release rate involves an increase in the physical
extent of the fire. For example, hydrocarbon pool fires that burn at around
2.25 MW/m2 [27-30] do not obligingly burn at a higher rate (without changing the
fuel type). To double the heat release rate, the pool area (area of the fuel source)
also needs to be doubled. That has the effect of keeping the temperature in the hot
plume relevant for the backlayering nearly constant with increasing HRR. A more
intense fire (greater heat release rate per unit area) would result in a hotter plume
relevant for the backlayering and greater propensity to backlayer. Experiments with
a fixed burner area and varying gas flow are actually varying the intensity of the
fire, introducing a variation that is not there for large real fires, yet the resulting
formulae do not control for fire intensity.
The third problem is the extent of the fire, and of course, it is related to intensity.
For a 20 m long fire (e.g. 20 m long truck on fire or 20 m long pool fire), the most
upstream part of the fire will have the most chance of contributing to backlayering,
10

with subsequent parts being progressively less influential. It seems, from CFD in-
vestigations, that heat released more than one tunnel height or so beyond the fire
front has less additional influence on the backlayering propensity (see Figure 15).
So, a 20 m long fire (e.g. truck or pool fire) uniformly burning at 100 MW would
have a similar critical velocity value as a 10 m long fire burning at 50 MW, and
probably not too much different to that for a 5 m long fire burning at 25 MW, when
assuming that the width of the fire stays the same. Especially for wide fires, it seems
that the downstream part of the fire after the first few metres does not contribute
much to the required critical velocity value (as is seen in the Memorial Tunnel test
results [27-30]). The heat release rate in those tests was increased from 10 to
100 MW by keeping the width of the fire the same but adding fire pans in the lon-
gitudinal direction. Consequently, the critical velocity value did not change very
much from 25 MW to 100 MW (see Figure 16).
The fourth problem is the effect of tunnel aspect ratio. There are a couple of
proposed corrections to formulae to account for the ratio of the tunnel width to the
height [12], [19-20]. The reason that such corrections are required is clear. The
recent (since about 1990) prior formulae all inherently use a temperature rise calcu-
lated by diluting the heat released by the fire into the entire flow along the tunnel
(e.g. see equations (6), (7) and (9)). It is obvious that for most tunnel fires, only a
part of the tunnel flow enters the fire plume in the first few metres of the fire (where
backlayering will be determined). As a tunnel gets wider, that fraction of the flow
entering the plume goes down (see Section 4.4). Starting the formulation with an
inappropriate temperature rise and then seeking to correct the answer empirically
based on geometry does not seem the best approach. As part of this work, a more
realistic basis for buoyancy, roughly modelling the plume area, is sought. Natural
break points then enter the formulation when the tunnel is so narrow that the plume
fills most of the cross section.
The next section seeks to develop a new, physically reasonable model, which
addresses the above shortcomings.

3 The Present Mixed-Convection Model

All the transport processes in tunnel fire events are governed by the equations of
motion of continuum fluid mechanics. These are the balance equations for mass,
momentum and thermal energy. The fluid motion contributes convective transport
of the conserved quantities. Tunnel fires inherently include both forced and natural
convection, where the latter is due to Archimedian buoyancy caused by the temper-
ature (or rather density) differences in the flow field. The occurrence of both types
of convection at similar strengths produces what is called mixed convection. This
is particularly the case when analysing the equilibrium where the buoyancy-driven
convection upstream of the fire (natural convection) interacts with the inertial force
of forced convection. The governing non-dimensional equations of motion are for-
mulated in a Cartesian coordinate system, with x 1 (x) aligned with the tunnel axis
in the flow direction, x2 (y) horizontal, and x3 (z) pointing in the direction of the
tunnel height. For the present flow regime (steady state, Newtonian fluid and Mach
number M < 0.3), the continuity equation for the incompressible fluid, the momen-
tum equation and the thermal energy equation are given as
𝜕𝑢𝑖∗
=0 (16)
𝜕𝑥𝑖∗
 11

𝜕𝑢𝑖∗ 𝜕𝑝 ∗ 1 𝜕 2 𝑢𝑖∗ 𝐺𝑟𝑖 ∗

𝑢𝑗∗ ∗ =− ∗+ − Θ (17)
𝜕𝑥𝑗 𝜕𝑥𝑖 𝑅𝑒 𝜕𝑥𝑗∗2 𝑅𝑒 2

𝜕Θ∗ 𝜕𝑝∗ 1 𝜕 2 Θ∗ 𝐸𝑐 ∗
𝑢𝑗∗ ∗
∗ = 𝐸𝑐 𝑢𝑗 ∗+ + Φ , (18)
𝜕𝑥𝑗 𝜕𝑥𝑗 𝑅𝑒𝑃𝑟 𝜕𝑥𝑗∗ 𝜕𝑥𝑗∗ 𝑅𝑒
respectively. The non-dimensional variables and unknowns are defined with the
length scale 𝐿, reference velocity 𝑈 , the coordinates 𝑥𝑖 and velocities 𝑢𝑖 . The
non-dimensional pressure 𝑝 ∗ = 𝑝/(𝜌𝑈 2 ), and the non-dimensional temperature is
defined by Θ∗ = (𝑇 − 𝑇𝑎 )⁄∆𝑇. The thermal buoyancy caused by density differ-
ences owing to temperature gradients 𝜌𝑔𝑖 𝛽∆𝑇 is expressed via the body force in
the momentum equation. This replaces the Froude number by the ratio 𝐺𝑟𝑖 ⁄𝑅𝑒 2 .
That is, the Froude number does not exist in the governing equations for mixed
convection and hence should not have been central to scaling of such flow regimes
as discussed in Section 2.3. The non-dimensional numbers 𝑅𝑒, 𝐺𝑟𝑖 , 𝐸𝑐 and 𝑃𝑟 are
the Reynolds, Grashof, Eckert and Prandtl numbers, respectively, defined for this
problem as
𝑈𝐿
𝑅𝑒 = (19)
𝜈𝑎

𝑔𝑖 𝛽𝐿3 ∆𝑇
𝐺𝑟𝑖 = (20)
𝜈𝑎2

𝑈2
𝐸𝑐 = (21)
𝑐𝑝 Δ𝑇

𝜇𝑐𝑝
𝑃𝑟 = (22)
𝑘
The Reynolds number is the ratio of inertial to viscous forces, where 𝑈 is the
characteristic velocity, 𝐿 the characteristic length of the flow field and 𝜈𝑎 the kine-
matic viscosity of ambient air. In the current flow problem, the hydraulic tunnel
diameter 𝐷ℎ will be adopted as the characteristic length 𝐿 in the Reynolds number.
The Grashof number is the ratio of buoyancy to viscous forces. As the buoyancy is
driven by the hot plume, the kinematic viscosity of ambient air 𝜈𝑎 becomes the kin-
ematic viscosity of the plume 𝜈𝑝 . For an ideal gas, 𝛽 = 1/𝑇. The characteristic
length 𝐿 is defined here as being the tunnel height from the base of the fire to highest
point of the ceiling 𝐿𝑛 . The term 𝑔𝑖 represents the gravitational acceleration com-
ponent in the Cartesian coordinate direction 𝑖. For tunnels with a moderate (typical)
longitudinal slope 𝑠 and a transverse slope 𝑠𝑡 , 𝑔𝑖 can be approximated as:
𝑔1 = 𝑔𝑥 ≅ 𝑔 ∙ 𝑠; 𝑔2 = 𝑔𝑦 ≅ 𝑔 ∙ 𝑠𝑡 ; 𝑔3 = 𝑔𝑧 ≅ 𝑔 (23)
That is, 𝐺𝑟𝑖 in the x and y directions becomes very small so that the body force
in these coordinate directions can be neglected for the current problem (strength of
forced convection dominates the strength of natural convection). In a further sim-
plification, 𝐺𝑟𝑖 in the z direction will be declared just as 𝐺𝑟.
The Eckert number is the ratio of the kinetic energy to the enthalpy difference
due to the temperature increase, and the Prandtl number represents the ratio of the
diffusivities for momentum and thermal energy of the fluid. In flow with strong
enthalpy differences as compared to kinetic energy, such as the present one, the
Eckert number is not important for the physical analysis of the flow. Yet, the
Prandtl number appears in a product with the Reynolds number in front of the heat
12

conduction term in equation (18). In flow dominated by convection, heat conduc-

tion is of minor importance, so that the Prandtl number does not appear as a quantity
relevant for the flow modelling either.
When assuming that the Reynolds number is high (say Re > 105), the extra stress
term with 1/Re in front must be modelled for turbulent flow. The ratio of Gr and
Re2 expresses the strength of natural convection relative to the strength of forced
convection. It may be of the order of unity for mixed convection processes. The
ratio 𝐺𝑟 ⁄𝑅𝑒 2 is also known as the Richardson number Ri as per
𝐺𝑟 𝑠𝑡𝑟𝑒𝑛𝑔𝑡ℎ 𝑜𝑓 𝑛𝑎𝑡𝑢𝑟𝑎𝑙 𝑐𝑜𝑛𝑣𝑒𝑐𝑡𝑖𝑜𝑛
𝑅𝑖 = 2
= (24)
𝑅𝑒 𝑠𝑡𝑟𝑒𝑛𝑔𝑡ℎ 𝑜𝑓 𝑓𝑜𝑟𝑐𝑒𝑑 𝑐𝑜𝑛𝑣𝑒𝑐𝑡𝑖𝑜𝑛
As noted above (Equation (5)) this has been referred to by prior authors as a
“critical Froude number” (not to be confused with the Froude number that has been
used in attempting to scale mixed convection problems), but is here referred to in
standard fluid mechanics terminology.
The gas temperature 𝑇 in the relation 𝛽 = 1/𝑇 can be expressed as the sum of
the approaching air temperature and the temperature increase due to the fire, i.e.
𝑇 = 𝑇𝑎 + ∆𝑇 (25)
Based on the relations above, the hydraulic diameter of the tunnel cross section
is selected as the length scale L in the Reynolds number. The physical flow pro-
cesses, however, are influenced by other length scales also, which enter via bound-
ary conditions. For the modelling of the mixed convective flow effects of the pre-
sent flow, therefore, the Grashof number for the z direction is modified by
incorporating a length scale ratio and a grade factor for sloped tunnels. The Reyn-
olds and modified Grashof numbers are:
𝑈𝐷ℎ
𝑅𝑒 = (26)
𝜈𝑎

𝐿 3 𝐿 3 𝑔𝐷ℎ3 ∆𝑇
̃ = 𝐾𝑔2 ( 𝑛 ) 𝐺𝑟 = 𝐾𝑔2 ( 𝑛 )
𝐺𝑟 (27)
𝐷ℎ 𝐷ℎ 𝜈𝑝2 𝑇𝑎 + ∆𝑇
The grade factor for sloped tunnels 𝐾𝑔 is defined as being the ratio of the critical
velocity in a sloped tunnel to the critical velocity in a flat tunnel, with the fire sce-
nario and other geometrical tunnel parameters kept the same (see Table 5).
𝑈𝑐
𝐾𝑔 = (28)
𝑈𝑐_𝑓𝑙𝑎𝑡
In this work, all transverse gradient (crossfall) has been taken as zero (𝑠𝑡 = 0 and
𝑔2 = 0). Large crossfall may warrant an additional correction parameter.
Formulating the Richardson number with the Reynolds number and the modified
Grashof number as defined by equations (26) and (27) and solving for the velocity
𝑈 gives:
1 ⁄2
𝑔𝐿3𝑛 1 𝑣𝑎2 ∆𝑇
𝑈 = 𝐾𝑔 ( 2 ) (29)
𝐷ℎ 𝑅𝑖 𝑣𝑝2 𝑇𝑎 + ∆𝑇
The velocity 𝑈 is related to smoke propagation, but is not critical velocity until 𝑅𝑖
and any empirical calibration is resolved for the equilibrium of the smoke layer at
the front of the fire site. Equation (29) suggests a dependence of the smoke control
related velocity on the kinematic viscosity of the gas. Since the flow regime
analysed is turbulent, the kinematic viscosity relevant for the flow is the turbulent
or eddy viscosity, which is orders of magnitude higher than the molecular kinematic
 13

viscosity of the gas. Even if there is a big difference in the dynamic viscosities of
the plume and the approaching air, the effective viscosity (i.e., the sum of the
molecular and turbulent viscosities) of the two regimes is very similar. The
viscosity ratio in equation (29) is therefore dominated by the eddy viscosities, with
similar values of 𝜈𝑎 and 𝜈𝑝 . This being the case, the viscosity ratio in (29) will be
close to unity, thus simplifying the equation. To check this simplifying approxima-
tion, Figure 1 shows results from simulations (specified in Section 4) carried out
with either a constant dynamic viscosity or a temperature-dependent dynamic vis-
cosity. The results regarding smoke propagation are, as expected, very similar and
confirm the conclusion.

(a) (b)

Figure 1. Simulation results of validation case 2 according to Table 1 in [51], with constant vis-
cosity (top) and temperature-dependent viscosity (bottom). (a) Contour plot of dynamic viscos-
ity in a vertical plane through the tunnel centreline. (b) Contour plot of gas temperature (in Kel-
vin) in a vertical plane through the tunnel centreline. Temperature and viscosity are clipped to
400°C and 4×10-5 Pa·s, respectively, to better visualise the differences in the upstream tempera-
ture layer. The clipped area is shown in magenta.

As already discussed, it is presumed that, at the critical velocity, the strength of

forced convection is of the same order as the strength of natural convection. Ac-
cording to (24), this requires that the Richardson number is close to unity.

The frontal area of the rising plume represents an obstacle for the flow past the
fire. If the plume frontal area is small compared to the tunnel area, a big fraction of
the flow goes around the plume, so that the momentum acting onto the plume is
lower than it would be for a plume frontal area that is nearly as big as the tunnel
area (wide fires). The plume in the latter case becomes more deflected by the up-
stream flow, and more mixing into the plume occurs, so that the upstream velocity
required to prevent backlayering goes down. The influence of the plume width on
critical velocity can be considered by the factor 𝐾𝐹 defined here as:
𝑊𝑓𝑖𝑟𝑒 𝐿𝑛
𝐾𝐹 = 𝑓𝑔 ∙ 𝑎𝑤 [1 − min (1, )] + 1 − 𝑓𝑔 (30)
𝐴
This correlation is a function of the plume frontal area 𝑊𝑓𝑖𝑟𝑒 𝐿𝑛 in relation to the
tunnel cross-sectional area 𝐴. The term ‘min(1, 𝑊𝑓𝑖𝑟𝑒 𝐿𝑛 / 𝐴)’ makes sure that the
plume frontal area does not exceed the tunnel cross-sectional area. The parameters
𝑓𝑔 and 𝑎𝑤 are empirical constants that will be calibrated to measurements and a
validated CFD model.

It is assumed that the effective temperature difference ∆𝑇 is a function of the

fire intensity (heat release rate by unit area), and the cooling due to the approaching
air. The higher the approaching air mass flow and the bigger the plume frontal area,
14

the more the hot plume gases are mixed and cooled down. The higher the fire in-
tensity, the higher is the density deficit and thus the buoyancy force opposing the
forced convection at the initial plume regime (fire front). It is also acknowledged
that an increase in the heat release rate caused by extending the fire source in the
longitudinal direction contributes less than the first part of the fire to the strength of
natural convection at the front of the fire. This reasoning leads to the following
representation of the effective temperature difference:
min(𝑄̇, 𝐼𝑓𝑖𝑟𝑒 𝑊𝑓𝑖𝑟𝑒 𝐿𝑛 𝑓ℎ 𝐾𝐿 )(1 − 𝜀)(1 − 𝜂)
∆𝑇 = (31)
min(𝐴, 𝐿𝑛 (𝑊𝑓𝑖𝑟𝑒 + 𝑚𝐿𝑛 ))𝑈 𝜌𝑎 𝑐𝑝
where 𝑄̇ is the total fire heat release rate, 𝐼𝑓𝑖𝑟𝑒 is the fire intensity (total fire heat
release rate per unit area), 𝑊𝑓𝑖𝑟𝑒 the width of the fire, 𝐿𝑛 the tunnel height from
the base of the fire to highest point of the ceiling, 𝜖 is the heat reduction due to im-
perfect combustion and/or radiation, 𝐴 is the tunnel cross-sectional area, 𝑈 is the
air velocity onto the fire, and 𝑓ℎ 𝐾𝐿 is describing the fraction of the length of the
fire (as a proportion of 𝐿𝑛 ) that contributes to the natural convection at the front of
the fire. The chosen function for the effective fire length 𝐾𝐿 is based on the ratio
of the longitudinal extent of the fire 𝐿𝑓𝑖𝑟𝑒 = 𝑄̇/( 𝑊𝑓𝑖𝑟𝑒 𝐼𝑓𝑖𝑟𝑒 ) (see equation (34) be-
low) to the tunnel height from the base of the fire 𝐿𝑛 and the empirical constants
𝑓𝑎 , 𝑓𝑜 and 𝛾𝐿 as follows:
𝑓𝑎 + 𝑓𝑜
𝐾𝐿 = 𝐿𝑓𝑖𝑟𝑒 (32)
−
𝑓𝑎 + 𝑒 𝛾𝐿 ∙𝐿𝑛

It was chosen simply as it has the desired form, and is able to be calibrated to
observations.
Taking the minimum of 𝑄̇ and 𝐼𝑓𝑖𝑟𝑒 times the effective fire area 𝑊𝑓𝑖𝑟𝑒 𝐿𝑛 𝑓ℎ 𝐾𝐿 en-
sures that the modelled effective heat release rate does not exceed the actual total
fire heat release rate if 𝐿𝑛 𝑓ℎ 𝐾𝐿 is larger than 𝐿𝑓𝑖𝑟𝑒 . The plume area interacting
with the approaching air is taken as 𝐿𝑛 (𝑊𝑓𝑖𝑟𝑒 + 𝑚𝐿𝑛 ). The basis for this is an as-
sumption that plume shapes will be geometrically similar, growing in width with
grade 𝑚 as the plume rises. As the plume width increases with height, the plume
frontal area interacting with the ambient air will then vary with height from the
base of the fire (conservatively tunnel height for design purposes). If a tunnel is
wider than the mean plume width, the additional air passing the fire to the side of
the fire does not dilute the effective temperature difference and therefore does not
decrease the buoyancy force. However, the plume frontal area cannot be bigger
than the tunnel area itself, which requires that the minimum of 𝐴 and
𝐿𝑛 (𝑊𝑓𝑖𝑟𝑒 + 𝑚𝐿𝑛 ) be taken as representing the plume frontal area interacting with
the ambient air. Deluge or water mist systems reduce the convective heat release
rate and therefore the effective fire intensity, which can be considered via the pa-
rameter 𝜂.

With the definition of the fire intensity

𝑄̇
𝐼𝑓𝑖𝑟𝑒 = (33)
𝑊𝑓𝑖𝑟𝑒 𝐿𝑓𝑖𝑟𝑒
the longitudinal extent of the fire can also be expressed as
𝑄̇
𝐿𝑓𝑖𝑟𝑒 = (34)
𝑊𝑓𝑖𝑟𝑒 𝐼𝑓𝑖𝑟𝑒
where 𝑄̇ is the total fire heat release rate.
 15

The definitions and assumptions above, finally lead, together with equations (29)
and (32), to the following system of implicit equations for the critical velocity
1⁄2
𝑔𝐿3𝑛 ∆𝑇
𝑈𝑐 = 𝐾𝐹 𝐾𝑔 ( ) (35)
𝐷ℎ2 𝑇𝑎 + ∆𝑇

min(𝑄̇ , 𝐼𝑓𝑖𝑟𝑒 𝑊𝑓𝑖𝑟𝑒 𝐿𝑛 𝑓ℎ 𝐾𝐿 )(1 − 𝜀)(1 − 𝜂)

∆𝑇 = (36)
min (𝐴, 𝐿𝑛 (𝑊𝑓𝑖𝑟𝑒 + 𝑚𝐿𝑛 )) 𝑈𝑐 𝜌𝑎 𝑐𝑝

For flat tunnels, the grade factor 𝐾𝑔 is assumed to be unity. The characteristic
length scale 𝐿𝑛 is defined as the height from the bottom of the heat source to the
tunnel ceiling (conservatively the tunnel height for design purposes), 𝐷ℎ is the tun-
nel hydraulic diameter, and 𝑊𝑓𝑖𝑟𝑒 is the fire width (most likely between 2.0 and
3.5 m for road tunnels). The fire intensity 𝐼𝑓𝑖𝑟𝑒 depends on the type of the fire and
the fire scenario. A typical value for a pool fire is 2.25 MW/m2 [27] and is con-
sidered as a conservatively intense fire scenario. A higher fire intensity might be
possible with different fuels or, in the experimental case, with burners introducing
fuel at high rates in small areas. The empirical factors and constants in the semi-
empirical correlation for calculating critical velocity (𝑓ℎ , 𝑓𝑎 , 𝑓𝑜 , 𝛾𝐿 , 𝑓𝑔 , 𝑎𝑤 and 𝑚 )
might be variable, depending on the fire characteristics. For pool fires, it has been
found that the values provided in Table 2 accurately represent the physical phe-
nomena (see Section 5). These values were first selected as fitting the Memorial
Tunnel test data and results from a validated CFD model (see Section 4), and were
subsequently found to require no alteration to fit with the available small scale test
data. The validity of the parameters listed in Table 2 is limited to the cases ana-
lysed or validated against (see Section 4 and 5). However, with no test data con-
flicting with the empirical results, they likely have wider applicability. How wide,
is subject to further investigations. If water mist or deluge systems are not consid-
ered, 𝜂 = 0. Values for 𝜀 usually lie between 0.2 and 0.4 [54].

Table 2. Empirical factors and constants used in the derived equations for calculating critical
velocity. Values are derived from validations against pool fires and results from a validated CFD
model (see Section 4 and 5).
𝑓ℎ 𝑓𝑎 𝑓𝑜 𝛾𝐿 𝑓𝑔 𝑎𝑤 𝑚
0.176 0.95 0.88 2.20 0.58 1.82 0.35

4 CFD Validation, Grade Factor and Parameter Variation

To properly calibrate equations (35) and (36) together with (29) and (32) against
the Memorial Tunnel data [27-29] it is required to understand the grade factor, as
the Memorial Tunnel had a 3.2% downgrade. There are no critical velocity data
from full size tunnels with varying grade, and scale models are not considered ap-
propriate for the purpose. Further, the available data in the literature [6, 14] provide
very different answers regarding the slope effect. It cannot be reliably said if one
or none of them are correct. To reference the results back to a flat tunnel requires
an analysis of the grade factor 𝐾𝑔 between a flat tunnel and the 3.2% downgrade.
Clearly the same information on grade effects will also be useful in design of tunnel
ventilation. For this, a CFD study using the simulation software ANSYS Fluent
16

version 2021 R1 [31-33] has been carried out. The first task was to carefully vali-
date the CFD technique against Memorial Tunnel data, to be confident that simula-
tions of slope effect would be meaningful. That validation has already been done
and was documented in a separate paper by Beyer and Stacey [51]. A short sum-
mary of the different fire cases and results will be provided here (Section 4.1 and
4.2). Table 3 provides an overview of the simulation methodology and boundary
conditions used. More comprehensive description of the Memorial Tunnel tests,
the CFD model including all boundary conditions, and the methodology used, can
be found in the work by Beyer and Stacey [51]. The CFD model validated for the
Memorial Tunnel was then used to analyse the same model but with different tunnel
slopes (Section 4.3). In a further step, a few parameters like width of fire pan, shape
of tunnel profile and area, longitudinal extent of fire, and fire intensity were adapted
and the results analysed to test the derived equations for a broader range of param-
eters (Section 4.4). Changes to the original validated CFD model, and the results
of the additional CFD study, are documented here.

4.1 Validation Cases

The Memorial Tunnel has a total length of about 850 m and a typical tunnel pro-
file with a curved ceiling (see Figure 4). Jet fans were installed in banks of three,
up- and downstream of the fire, to adjust the tunnel air velocity during the fire tests.
Instrument trees (“loops”) were located adjacent to the fire and further up- and
downstream to measure velocity and temperature profiles and further flow parame-
ters along the tunnel. The measurement equipment and data acquisition units cre-
ated a fair blockage and were therefore implemented in the CFD model, based on
sketches and the reported blockage area. Location of the measurement loops, as
well as of the individual sensors in the tunnel profile, are shown in Figure 3.
The Memorial Tunnel fire tests included pool fires with nominal heat release
rates (HRR) of 10, 20, 50 and 100 MW. Each test used the appropriate fire pan, or
a combination of the fire pans, as illustrated in Figure 2 and listed in Table 4. To
validate the CFD model for low as well as high heat release rates, one test with a
nominal HRR between 10 MW and 20 MW, one at 50 MW, and a third one at
100 MW were selected. The CFD methodology and model boundary conditions are
listed in Table 3. Essential case parameters are listed in Table 2. The flow param-
eters and measured HRR values listed are averages over a period when the flow and
smoke layer were fairly constant. This allows the comparison of the measured val-
ues with a steady-state simulation in a reasonable way. More details about the ge-
ometry, test setup and simulation methodology are provided by Beyer and Stacey
[51].
 17

Figure 2. Computational mesh around the fire pans and adjacent instrument loops.

Table 3. CFD methodology and boundary conditions for validation cases.

Pressure based, steady-state, Reynolds-averaged
Solver Navier-Stokes (RANS) with spatial discretization
method of 2nd order.
Turbulence Realizable k-ε model with scalable wall function.
Incompressible ideal gas law. Piecewise-polyno-
mial function for the temperature dependent spe-
cific heat capacity for the individual species. Ther-
mal dependency of the thermal conductivity and
Thermal effects
the dynamic viscosity for the individual species
were estimated based on Sutherland’s law with
three coefficients, in combination with a mass-
weighted mixing law.
Representation of fire Eddy-dissipation combustion model.
Not simulated. Instead, a radiative fraction ap-
proach as explained in the literature referenced [35,
Thermal radiation
51] has been adopted, capturing the loss of thermal
energy in a gross sense.
Ideal combustion of fuel oil with a chemical for-
mula of C19H30. Additional mass source term for
Fuel
ideal combustion and the radiative fraction ap-
proach (see [51]) are incorporated.
Mass flow inlet with air temperature that was
measured upstream of fire (~Loop 207). Mass
flow is based on Loop 214 (mass fraction of
Tunnel inlet
O2 = 0.23, N2 = 0.77 and 0 for all other species).
See Table 4 and Beyer & Stacey [51] for further
details.
Tunnel outlet Outflow
Wall temperature is ambient air temperature just
upstream of fire (~Loop 207), rough wall with
Tunnel wall
roughness height = 0.0015 m. See Table 4 and
Beyer & Stacey [51] for further details.
Wall – instrument trees,
Adiabatic, rough wall with roughness
obstacles, jet fans, fire
height = 0.00025 m
pan frames
18

Fan boundary to implement flow rate of 42.95 m3/s

Jet fans and discharge velocity of 34.2 m/s for activated jet
fans according to Table 4.
Mass flow inlet with ambient temperature
(~Loop 207) on surface located on top of fire pan.
Mass fraction of O2 and N2 was 0, mass fraction of
fuel and other species according to adopted fire
Fire pan
HRR, radiative fraction and combustion efficiency.
Combustion efficiency for extra mass source is as-
sumed to be 95% [27, 28]. See Table 4 and
Beyer & Stacey [51] for further details.

Table 4. Input and test parameters extracted from the Memorial Tunnel test data [27-29] for
three validation cases as documented in Beyer & Stacey [51]. Bulk temperatures are mass-aver-
aged temperatures.
Parameter Case 1 Case 2 Case 3
Test number 606A 610 615B
1651 to 191.5 to 678.5 to
Elapsed time used for averaging (sec.)
1680.5 281 828.5
Measured total HRR1(MW) 11.1 54.3 104.6
JF2, 5, 8, JF1, 3, 4,
Activated jet fan/s JF3
11 & 14 6, 7 & 9
50, 30 &
Fire pan/s used, according to Figure 2 10 MW 50 MW
20 MW
Flow rate - Loop 214 (m³/s) 143.2 148.3 139.8
Average velocity - Loop 305 (m/s) 2.92 3.00 2.85
Blockage area at Loop 3053 (m2) [27], [29] 10.2 10.2 10.2
Free tunnel cross sectional area (m2)4 59.9 59.9 59.9
Average velocity unobstructed tunnel section (m/s) 2.44 2.51 2.38
Upstream temperature - Loop 214 (°C) 3.6 4.0 6.8
Upstream temperature - Loop 207 (°C) 8.5 7.3 12.0
Upstream temperature - Loop 307 (°C) 9.0 7.2 14.8
Measured bulk temperature after fire - Loop 302 (°C) 53 190 451
Measured bulk temperature after fire - Loop 303 (°C) 57 217 475
Upstream density - Loop 214 (kg/m³) 1.28 1.27 1.26
Specific isobaric heat capacity (kJ/kgK) 1.014 1.044 1.096
Theoretical bulk temperature after the fire based on to-
69 282 554
tal HRR(°C)
Radiative fraction2 (-) 0.21 0.24 0.17
1
Measured heat release rate corrected by the CO to CO2 ratio to account for non-ideal combus-
tion [27], [28]
2
Ratio between measured bulk temperature and the theoretical bulk temperature rise resulting
from the total HRR is assumed to be the radiative fraction.
3
Value as stated in Section 8.8.4 in the comprehensive test report [27], and in the paper from
Kile & Gonzalez [29]. The fraction of the tunnel area blocked is about 17%, as also reported in
the same references.
4
Tunnel height and free tunnel cross sectional area are calculated assuming that the removed
ceiling had been flat. It may in fact have had a slight camber (see Figure 4) such that the true
height and area inferred from the dimensions were slightly higher than tabulated (possibly 1%).
 19

19.8 m 86.0 m 105.2 m 110.0 m 105.5 m 82.0 m 45.1 m 61.9 m 66.1 m 41.8 m 110.6 m 19.8 m

41.8 m

32.6 m 18.0 m 11.3 m 17.4 m 36.6 m

12.2 m

Figure 3. Instrument loop locations, with instrument types at each loop. Sketch from the Memo-
rial Tunnel report [27-28].

Figure 4. Memorial Tunnel cross section with dimensions, looking north [27-28]. The dimen-
sioning did not include what may have been a slight camber in the ceiling, which if included,
would increase tunnel height and area by perhaps 1% compared to the values used here.

4.2 Validation Results and Discussion

For the three simulated validation cases listed in Table 4, Figure 5 to Figure 7
compare the predicted bulk temperature along the tunnel axis and the temperature
profiles over the tunnel height against the corresponding test results [27-28]. Mul-
tiple thermocouple readings at a given elevation are averaged in the temperature
profiles at the different measurement loops. The representative elevation area is a
20

horizontal slice of the tunnel cross-section around the elevation instruments as spec-
ified in the test reports [27-28]. Identical horizontal slices have been used for eval-
uating the CFD simulations. For a better understanding of the temperature profiles,
a contour plot of the temperature through the middle of the tunnel around the vicin-
ity of the fire is also given for each case.
In all cases, the temperature profiles downstream of the fire were predicted
with reasonable accuracy with the applied simulation methodology. The higher
temperature readings of the test data at the upper part of the temperature profile
close to the fire (Loops 304 and 303) might be related to the influence of thermal
radiation on the sensors (see Figure 5 and Figure 6) as noted in the test reports
[27-28].
Besides the deviation of the bulk temperature close to the fire (influence due to
thermal radiation on the temperature sensors during the tests), the bulk tempera-
ture at Loop 202 also shows a slight deviation, especially at higher downstream
temperatures (higher HRR). Loop 202 is located in the smaller cross section near
the south portal with the reduced ceiling height [51]. The step change in the cross-
section area is just 1.5 m upstream of Loop 202. This step will have created a sep-
aration zone and recirculation close to the ceiling. The difference in bulk tempera-
ture seen at Loop 202 may be related to the difficulty in turning point measure-
ments of velocity and temperature into bulk averages, for such a non-uniform
flow.
The velocity in the tunnel during the Memorial Tunnel tests was adjusted using
the jet fans installed in the tunnel upstream of the fire. These jet fans were not
speed controlled. To change the air speed in the tunnel, a jet fan could be
switched on or off, making the speed control fairly coarse. The buoyancy force
and flow resistance of the 11 MW fire are quite low, certainly much lower than for
a 50 MW or 100 MW fire. With the coarse control on jet fan forcing, the low re-
sistance made it difficult to achieve a steady velocity where backlayering was just
prevented. For the tests with low heat release rates, two active jet fans were insuf-
ficient to stop smoke propagation upstream of the fire, and three jet fans were too
strong, so that the approaching air velocity was higher than required for prevent-
ing backlayering. The point at which backlayering was just prevented was ana-
lysed during the velocity change after turning the third jet fan on or off. This
made it difficult to accurately determine the critical velocity for low heat release
rates. The scatter in the reported data at low HRR reflects this (see Figure 16).
As discussed in detail in Beyer and Stacey [51], the results below are evaluated
at times when smoke upstream of the fire was recently moved downstream and
reached almost steady state conditions for a short period. Some of the temperature
readings upstream of the fire show slightly increased values which may possibly be
explained by the thermal inertia of the temperature sensors only recently out of the
smoke layer. Especially for the lower HRR cases, Loop 304 just a few metres
downstream of the fire pans gives higher temperature readings in the upper part of
the tunnel than predicted. This may be related to sensor hardware thermal inertia
and a transient behaviour of the flames and hot plume moving up and down and
heating up the sensors in the upper part of the tunnel. A more comprehensive dis-
cussion about the smoke propagation during the tests and the deviations of the tem-
perature profiles for validation cases is given in Beyer and Stacey [51].
 21

(a)

(b)

(c)
Figure 5. 11 MW validation (Case 1 from Table 4) with combustion model. Comparison of
simulation results with test results for test 606A [51]: (a) Bulk temperature along the tunnel axis,
(b) Temperature profiles over the tunnel height for different loops in the vicinity of the fire, (c)
Contour plot of the temperature in °C through the middle of the tunnel and fire pans. Temperature
is clipped to 150°C for a better visualisation of the temperature layering.
 (a)

(b)

(c)
Figure 6. 54.3 MW validation (Case 2 from Table 4). Comparison of simulation results with
test results for test 610 [51]: (a) Bulk temperature along the tunnel axis, (b) Temperature profiles
over the tunnel height for different loops in the vicinity of the fire, (c) Contour plot of the temper-
ature in °C through the middle of the tunnel and fire pans. Temperature is clipped to 400°C for a
better visualisation of the temperature layering.
 23

(a)

(b)

(c)
Figure 7. 105 MW validation (Case 3 from Table 4). Comparison of simulation results with
test results for test 615B [51]: (a) Bulk temperature along the tunnel axis, (b) Temperature pro-
files over the tunnel height for different loops in the vicinity of the fire, (c) Contour plot of the
temperature in °C through the middle of the tunnel and fire pans. Temperature is clipped to
800°C for a better visualisation of the temperature layer.
24

Based on the observations provided by Beyer and Stacey [51] and above, and ac-
knowledging a few uncertainties in the temperature readings, the smoke propaga-
tion and temperature distribution can be predicted within a reasonable accuracy by
the simulation techniques and procedures applied. Most importantly for the pre-
sent purpose, the smoke extent matches the tests very well for the same condi-
tions. However, it is important to check other parameters, to see that the critical
velocity is not correct by an ‘accident’ of compensating errors that may not com-
pensate each other in other scenarios. The temperature profiles away from radia-
tion effects are also predicted quite well, giving confidence that the analysis is a
reasonable representation of the plume and smoke layer dynamics. That is; the
technique is a reasonable tool for exploring trends in critical velocity, around the
validated cases.
The trend of interest here is the effect of tunnel slope on the critical velocity, and
as noted at the start of the section, the CFD work reported above was all with the
primary purpose to obtain a validated simulation methodology for exploring the
tunnel slope influence, with a secondary objective to look at the influence of other
geometric and fire parameters. The outcomes are described in the following sec-
tions.

4.3 Grade Factor

The validated CFD-model described by Beyer & Stacey [51] and summarized
earlier was used to identify the grade factor 𝐾𝑔 as defined in Section 3. For this
purpose, the jet fans and all the instrument trees and other obstacles have been re-
moved from the model, first to reduce the complexity and computational expense,
and second, to simplify assessment of when critical velocity is achieved, which is
made more difficult with the obstructions. As the models with different slopes are
compared in a relative sense, these geometric modifications are not thought to be
significant, and assessing critical velocity consistently across all tunnel slopes was
seen as far more important. Grade factors have been analysed for a 50 MW fire
(validation case 2), which is believed to be the most relevant fire scenario for road
tunnels.
To accurately reproduce the point where upstream smoke propagation is just
prevented, a simple feedback controller for adjusting the upstream velocity was
implemented in the simulation model. Upstream velocity is automatically in-
creased or decreased so that the tip of an upstream smoke layer is confined at the
front edge of the fire pan. This was done by tracking the maximum occurring
temperature at the tunnel cross sectional area in front of the fire front (‘tracking’
area) and adjusting the approaching air velocity in increments proportional to the
temperature ratio (between maximum temperature at the ‘tracking’ area and the
approaching air temperature). Once the temperature at the tracking area is equal
to the ambient air temperature, the approaching air velocity was decreased by
small, fixed increments. The inlet velocity stays constant once the temperature
 25

difference between the ‘tracking’ area and ambient air is >10 K and <40 K. By al-
lowing sufficient computational time, the same criteria could be applied precisely
to each slope case, eliminating any human judgement based on CFD images, and
hence refining the relative grade effect. The outcome of that CFD study is sum-
marized in Table 5 below.

Table 5. Grade factor for different tunnel slopes based on a 50 MW fire.

Tunnel slope 1 0% -3.2% -6.0%
Heat release rate (MW) 54.3 54.3 54.3
Average upstream velocity (m/s) 3.006 3.037 3.050
Grade factor 𝐾𝑔 (-) 1.000 1.010 1.015
1
Negative value for downgrade.

The conclusion from Table 5 is that, for most design purposes, the grade factor
can be ignored. For the practical range of road tunnel longitudinal slopes, the ef-
fect of slope appears to be much smaller than tolerances in calculating the required
ventilation to achieve critical velocity, or in measuring the result on commission-
ing. That is; for practical purposes, the factor 𝐾𝑔 may be deleted from equation
(35). Eliminating grade factor is a significant departure from prior practice, but
the basis of the above assessment that grade effects are not significant for typical
road tunnels is considered more robust than the prior trends based on very small
models or tests with methane layering [6, 14]. It seems likely that no grade factor
is required for transverse gradient (crossfall), but that has not been confirmed.

4.4 Parameter Variation

Based on the validated CFD model and the presented simulation methodology
[51], geometric parameters like width of the fire, tunnel height, tunnel profile, lon-
gitudinal extent of the fire, as well as the fire intensity, were varied to better un-
derstand the effect of those parameters on the critical velocity. For most scenar-
ios, the same velocity/backlayering controller described in Section 4.3 was used
for the parameter variation study. The results in turn were used to verify the pre-
sented physical model (equations (29), (32) (35) and (36)) and the empirical pa-
rameters as listed in Table 2.

Fire Width

The first parameter to be varied was the width of the fire pan. Fire width effects
were studied based on the simplified Memorial Tunnel model that was used for
the grade factor study. The tunnel slope of -3.2% was kept the same for the differ-
ent scenarios. The original 50 MW fire pan according to Figure 2 was varied in
26

width as illustrated in Figure 8, maintaining the fire pan length of 6.1 m and the
fire intensity of 2.43 MW/m2 according to validation case 2 (see Table 4). The
upstream air temperature and therefore the wall temperature in the simulation was
6.13°C. Further parameters and results of the simulations are listed in Table 6.
As already discussed, the wider the fire, the greater is the interaction with the ap-
proaching air. This has two effects. First, the increased plume frontal area causes
more mixing and secondly, it provides less space for the upstream air to go around
the plume (higher momentum onto the plume frontal area). Both have the effect
of decreasing the required critical velocity. As an outcome of that study, it also
seems that the critical velocity does not change if the fire width (fire blockage)
goes below a specific value (here 2.5 m) when the fire intensity is kept the same.
The proposed physical model shows a reasonably good agreement with the sim-
ulation results but tends to underestimate critical velocity for very narrow fires by
about 3%. Table 6 and Figure 9 provide a comparison between the simulation re-
sults and the semi-empirical model.

Figure 8. Schematic of Memorial Tunnel profile used for fire pan width variation.

Table 6. Case parameter and simulation results for different fire pan widths.
Fire width m 1.50 2.50 3.66 6.00
Area ratio plume/tunnel# - 0.82 0.70 0.57 0.29
Total HRR MW 22.2 37.1 54.3 89.0
Fire intensity MW/m2 2.43 2.43 2.43 2.43
Radiative fraction - 0.27 0.27 0.27 0.27
Critical velocity (CFD) m/s 3.20 3.19 3.03 2.50
Grade factor - 1.01 1.01 1.01 1.01
Critical velocity obtained
from proposed physical m/s 3.11 3.16 3.03 2.50
model
#
Area ratio based on equation (30) and defined as 1 − 𝑊𝑓𝑖𝑟𝑒 𝐿𝑛 /𝐴
 27

Figure 9. Critical velocity values in the Memorial Tunnel profile for different fire pan widths,
but constant fire intensity and fire pan length obtained from the validated CFD model and the
semi-empirical equations (29), (32) (35) and (36).

Tunnel Height and Aspect Ratio

The influences of the tunnel height and the aspect ratio were explored for a flat
rectangular 3-lane tunnel with a width of 12 m. This was done by using the sim-
plified Memorial Tunnel model parameters, replacing the Memorial Tunnel profile
by a rectangular profile as depicted in Figure 10. The tunnel slope was also re-
moved. The rectangular pool fire surface with a length of 20 m and a width of
2 m was on the floor and located 245 m downstream of the entrance portal. Fire
intensity was 1.25 MW/m2 which resulted together with the fire surface in a total
HRR of 50 MW1. The upstream air temperature and wall temperature in the simu-
lation was set to 21.4°C. Only the tunnel height varied from 3 m up to 12 m but
all other parameters were kept the same. Further parameters and results of the
simulations are listed in Table 7 below.
The simulation showed that the critical velocity increases from a tunnel height of
3 m (aspect ratio of 1:4) to 12 m (aspect ratio of 1:1) by 55%.
The proposed physical model accurately predicts the trend of the critical velocity
with tunnel height and aspect ratio, but underestimates the values by up to 14%.
This may be related to the rectangular tunnel profile and the reduced velocity at
the corners. The momentum onto the fire and potential backlayering at those areas
is reduced so that the smoke starts to propagate upstream along the corners. That

1
The low fire intensity resulted in matching the approximated fire intensity of a
volumetric heat source (that was 20 m long and 2 m wide) conducted in a different
study (not documented here or relevant for this study), to allow comparison. Even
if the fire intensity used here is regarded as low, it is still useful as validation as the
proposed equations should also work with different fire intensities.
28

physical effect related to a rectangular tunnel profile is not captured in the pro-
posed equations. For practical purposes, it is suggested either to increase the re-
sulting values by 7% to 15% for rectangular tunnel profiles, or simply to accept
some backlayering in the corners. Table 7 and Figure 11 provide a comparison
between the simulation results and the semi-empirical model.

Figure 10. Schematic of the tunnel profile used for the tunnel height and aspect ratio varia-
tion.

Table 7. Case parameters and simulation results for different tunnel heights and aspect ratios.
Tunnel height m 3 6 9 12
Aspect ratio - 1:4 1:2 3:4 1:1
Fire width m 2.0 2.0 2.0 2.0
Total HRR MW 50.0 50.0 50.0 50.0
Fire intensity MW/m2 1.25 1.25 1.25 1.25
Radiative fraction - 0.35 0.35 0.35 0.35
Critical velocity (CFD) m/s 2.2 2.9 3.2 3.4
Grade factor - 1.0 1.0 1.0 1.0
Critical velocity obtained
from proposed physical m/s 2.0 2.5 2.8 3.2
model
 29

Figure 11. Critical velocity values in a 3-lane rectangular tunnel profile for different tunnel
heights, but constant fire dimensions and HRR obtained from the validated CFD model and the
semi-empirical equations (29), (32) (35) and (36).

Variation of Fire Dimensions

For this study, the Memorial Tunnel profile of the simplified validation model
used for the grade correction has been replaced with a similar profile but with
3 lanes as shown in Figure 12. The total tunnel length was shortened to 550 m and
is downgrade at 5%. For the pool fire, a rectangular surface was defined and
placed on the floor. The fire front is located 300 m downstream of the entrance
portal. To understand the influence of the fire dimensions, the width and length of
the fire source area was varied by keeping the fire intensity of 2.5 MW/m2 con-
stant (fuel mass flow per m2 was constant). Depending on the fire source area, the
constant fire intensity resulted in different total fire heat release rates. In a further
step, some of the cases were analyzed with a downgrade slope of 1.7%. The up-
stream air temperature and wall temperature in all simulation cases was set to
13.85°C. Further parameters and results of the simulations are listed in Table 8
below.
The results for the bigger tunnel profile suggest that there is no difference in crit-
ical velocity between 1.7% and 5% downgrade. These results support the conclu-
sion made in Section 4.3, that for practical purposes, the factor 𝐾𝑔 may be taken to
be unity. These simulations confirm that for the same fire intensity, a wider fire
reduces the required critical velocity but a fire more extended in longitudinal di-
rection increases the critical velocity as indicated by equation (32).
The effect of the variation of the fire dimensions is also captured accurately by
the derived physical model, especially if the grade correction is neglected (𝐾𝑔 =
1) as advocated. Only the case where the fire is wider than it is long shows an
overprediction (of 8%). That overprediction can be accepted as such a design fire
scenario may be an edge case anyway. Table 8 and Figure 13 provide a compari-
son between the simulation results and the semi-empirical model.
30

Figure 12. Schematic of the tunnel profile used for the fire width and length variation.

Table 8. Case parameter and simulation results for different fire widths and lengths.
Fire length m 8 32 4 16
Fire width m 2.5 2.5 5.0 5.0
Total HRR MW 50 200 50 200
Fire intensity MW/m2 2.5 2.5 2.5 2.5
Radiative fraction - 0.3 0.3 0.3 0.3
Critical velocity (CFD)
m/s 3.62 4.06 3.16 3.70
for 5.0% downgrade.
Critical velocity (CFD)
m/s 3.62 4.06 - -
for 1.7% downgrade
Critical velocity obtained
from proposed physical
m/s 3.67 4.05 3.41 3.62
model (no grade correc-
tion, 𝐾𝑔 = 1)
 31

Figure 13. Critical velocity values in a 3-lane TBM tunnel profile for different fire widths and
lengths as well as tunnel slopes, but constant fire intensity. Values were obtained from the vali-
dated CFD model and from the semi-empirical equations (29), (32) (35) and (36).

Variation of Fire Intensity and Fire Length

In a final step, the validated simulation methodology [51] was applied on a TBM
(Tunnel Boring Machine) tunnel profile with 3 lanes and a false ceiling (to under-
stand the effects of varying fire intensity and fire length on critical velocity. In
this case, the tunnel is flat and has a total length of 530 m. The rectangular fire
source with a constant width of 2.0 m was placed on the floor, with the front of the
fire source located 300 m downstream of the entrance portal. Fire intensity was
varied by keeping the total fire heat release rate constant with varying fire length.
In a second step, fire intensity was varied while keeping the fire length constant.
Note, this variation acknowledges that the critical velocity mainly depends on the
physical dimensions of the fire and the fire intensity and not on the fire heat re-
lease rate itself. The upstream air temperature and wall temperature in all simula-
tion cases was 20.0°C. Further parameters and results of the simulations are listed
in Table 9 below.
The results for this tunnel profile confirm that critical velocity increases strongly
with fire intensity for the same total heat release rate. As before, the critical ve-
locity increases slightly with the fire length (as modelled in equation (34)).
That behavior was also replicated with the proposed physical model, as listed in
Table 9 and illustrated in Figure 15. The maximum deviation between the CFD
and the proposed model is 3%.
32

Figure 14. Schematic of the tunnel profile used for the fire intensity and fire length variation.

Table 9. Case parameters and simulation results for different fire intensities and fire lengths.
Fire intensity MW/m2 0.625 2.5 2.5
Total HRR MW 50 50 200
Fire length m 40 10 40
Fire width m 2.0 2.0 2.0
Radiative fraction - 0.35 0.35 0.35
Critical velocity (CFD) m/s 2.20 3.10 3.28
Critical velocity obtained
from proposed physical m/s 2.14 2.94 3.26
model

Figure 15. Critical velocity values in a 3-lane TBM tunnel profile with a false ceiling, for dif-
ferent fire intensities and fire lengths but constant fire width, obtained from the validated CFD
model and the semi-empirical equations (29), (32) (35) and (36).
 33

5 Validation of Mixed-Convection Model Against Real Test

Data

The behaviour of the proposed physical model with varying parameters has been
shown above to agree quite closely with results obtained from a validated CFD
model. This section documents the agreement of the proposed model with critical
velocity values obtained from longitudinal ventilation full-scale test results of the
Memorial Tunnel Fire Ventilation Test Program [27], [28-30], and also the agree-
ment with values obtained from small-scale tests as recorded by several research-
ers [13], [15], [18]. This validation against the small-scale tests was done without
attempting any scaling, but simply by using the real dimensions of the test ducts,
as well as the real measured velocities and heat release rates. As discussed earlier,
there is no approach to scaling the formula that does not involve major assump-
tions about at least some of the important parameters. Using the real parameters
ensures that the physical phenomena are retained, and not ‘assumed away’ and
therefore allows a verification that the mathematical model captures the important
physics appropriately.

5.1 Memorial Tunnel Fire Ventilation Test Program

Figure 16 compares the critical velocity values from the Memorial Tunnel tests
[27], [28-30] with the values given by the derived physical model (equations (29),
(32) (35) and (36)). The scatter in the plotted data points produced from the pro-
posed equations is due to the fire intensity being calculated from the reported
HRR and the actual pan area. That is; they include some element of the experi-
mental scatter within the original Memorial Tunnel data. For the proposed model,
a fixed radiative fraction of 0.21 was adopted, as being the average observed from
the validation cases (see Table 4). The grade factor was set according to Table 5.
Apparently, there are two versions of the ASHRAE paper from Kile & Gonzalez
published in 1997 [29], [30]. As it is not clear which version presents the most re-
liable critical velocity values, both versions are plotted in Figure 16 ([29]: trian-
gles, [30]: circles). The data points from the first version [29] are the same as
published in the comprehensive test report [27]. However, the data points from
the second version [30], seem to be more aligned with the results presented by
Kennedy [6]. The difference in the critical velocity values for high HRRs
(>50 MW) is minor and probably less of a concern, but the difference is quite big
for low HRRs (8 to 16 MW).
As noted earlier, the air velocity in the Memorial Tunnel tests was adjusted by
jet fans without variable speed drives. The change in air velocity caused by start-
ing an additional jet fan or stopping a running jet fan was too large to allow an ac-
curate determination of the velocity where backlayering of hot smoke was just
34

prevented at low heat release rates. This is likely to be the reason for the high var-
iation of the reported critical velocity values at low heat release rates. The outlier
at low heat release rates (critical velocity > 3.2 m/s) was re-evaluated by looking
at the raw data and 3 m/s may be a more reasonable value (see squares in Fig-
ure 16).
With those notes on the Memorial Tunnel data, we return to the comparison
against the validated CFD model. The Memorial Tunnel results show that for the
50 MW fire (Case 2 in Table 4) an air velocity of 3 m/s onto the fire was sufficient
to prevent backlayering (see Figure 16). That is also consistent with the grade fac-
tor study as given by Table 5. The results of the 100 MW case (Case 3 in Table 4)
indicate that a velocity of 2.85 m/s was high enough to control upstream smoke
propagation (Figure 16) but was perhaps too low to completely avoid backlayering
(see Figure 7). A value of 3.0 to 3.1 m/s may be sufficient to prevent backlayering
for that 100 MW case. Also, the validation case 1 (Table 4 and Figure 5) shows
that the velocity of 2.9 m/s onto the 10 MW fire was higher than critical velocity.
Consequently, a value around that or slightly less for a 10 MW fire may be suffi-
cient. That is; across the range of HRR of the Memorial Tunnel tests, the pro-
posed mixed convection model is consistent with the original test results and the
CFD validated against those results.
When acknowledging some uncertainties in the published data for low HRR, as
discussed above, the agreement of the proposed physical model with the Memorial
Tunnel data in Figure 16 can be seen as acceptable.
Figure 16 includes the predicted critical velocity according to Kennedy [6] and
those versions of the 2014 NFPA 502 equation published in 2014 [44] and in 2023
[48] for comparison. The difference between the various versions is discussed in
Section 2.2.
 35

Figure 16. Critical velocity determined from the Memorial Tunnel tests (triangles, circles and
squares) compared to the mixed convection model (orange dots) and the equation proposed by
Kennedy and two ‘versions’ of NFPA 502 2014 [44, 48] (dotted and dashed lines).

5.2 Small-Scale Test Data

All but one of the test rigs giving results used for comparison below, used circu-
lar burners. The current formula was based on rectangular heat sources, with the
Memorial Tunnel tests in mind. For the comparison purposes, the circular small-
scale burners have been taken as equivalent rectangular burners as follows. From
the proposed model, the width is the most important dimension of the fire. Pro-
vided that the maximum width (the diameter) falls within the effective length,
which it does in all cases, the fire width can be taken as the burner diameter. In
order to preserve both the fire intensity and the HRR, the burner area needs to be
unchanged, so the fire length is then taken as the burner area divided by the width.
For circular burners of diameter 𝑑𝑏 , this makes the fire width 𝑑𝑏 , and the fire
length 𝐿𝑓𝑖𝑟𝑒 = 𝜋𝑑𝑏 /4.
Figure 17 and Figure 18 compare the critical velocity values from different
small-scale tests with the values predicted by our mixed convection model. As
previously discussed, and also elaborated by Stacey and Beyer [36], the fire inten-
sity is a measure for the density deficit, with the buoyancy force increasing with
36

the fire intensity. The burner size was kept the same during the referenced model
tests [13], [15], [18] (for the individual test setup), which means that the fire inten-
sity (MW/m²) in these models increases in proportion to the HRR. This is the rea-
son why the critical velocity in the scale model tunnels climbs steeply with in-
creasing HRR.
As discussed by Wu & Bakar [15], with further increase in HRR, the plume gets
elongated in the downstream direction and the critical velocity become independ-
ent of the HRR once the intermittent flame zone reaches the tunnel ceiling. That
is seen by the kink in the measured data, which can be observed earlier in the
smaller and more confined test tunnels. If the tunnel gets higher and wider, the
position of the kink moves towards higher heat release rates (see Figure 17 (a) vs.
Figure 17 (b)). In all small-scale tests considered here, the kink occurs when the
average temperature after the fire exceeds 330°C to 370°C.
This behaviour could be reproduced with the proposed model by restricting the
effective temperature difference ∆𝑇 in equation (35) once the average temperature
after the fire exceeded 330°C to 370°C (see green dashed lines in Figure 17 and
Figure 18). Whatever the precise location of the kink is, the mechanism is be-
lieved to be that the fire intensity becomes nearly constant, with the effective area
of combustion extended downstream in order to get all the injected fuel burnt. If
that is so, as indicated by Wu and Bakar [15], the proposed model predict a con-
stant critical velocity. However, it appears that this is an artefact of the burner
types used in the very confined space of small-scale tunnels, and therefore not so
relevant for real tunnel fires.
Oka and Atkinson [13] as well as Wu and Bakar [15] used a water spray device
to cool the tunnel walls near the fire source. As noted by Oka and Atkinson [13],
the wall temperature has significant impact on the critical velocity. This can also
be observed when comparing the test results of similar test tunnels (Tunnel A in
Figure 17 and Tunnel B in Figure 18). The cooling effect of the water spray sys-
tem could be reproduced with the proposed equations by considering a reduction
of the convective heat release rate (𝜂 = 30%) and hence further restricting the
temperature difference ∆𝑇 in equation (35) (see blue dotted lines in Figure 18).
The value of 𝜂 = 30% was derived by comparing Tunnel A in Figure 17 with
Tunnel B in Figure 18 and then kept the same for the wider Tunnels C (Figure 18
(c)) and D (Figure 18, (d)). For the small-scale tests conducted by Oka and Atkin-
son [13], a value of 𝜂 = 10% gave agreement when applied in the proposed equa-
tions.
Acknowledging the uncertainties around thermal restriction due to flame/burner
behaviour and the use of water cooling, our proposed model shows a reasonably
good agreement with small scale test results, as illustrated in Figure 17 to Fig-
ure 18, and therefore also represents the physical phenomena in small-scale tun-
nels adequately.
 37

(a) (b)

Figure 17. Critical velocity values measured in small scale tunnels as published by Li et al. [18]
compared to the proposed physical model and the trend line proposed by Li et al. [18]: (a) Test
tunnel A; (b) Test tunnel B.

(a) (b)

(c) (d)

Figure 18. Critical velocity values measured in small scale tunnels compared to the derived
physical model: (a) test results and trend line as proposed by Oka and Atkinson [13]; (b-d) test
results and trend line as proposed by Wu and Bakar [15] for different small-scale tunnel profiles.
38

6 Practical estimation of critical velocity

The proposed physical model is seen above to accurately model the Memorial
Tunnel outcomes, and also accurately follows small-scale test data until the
flame/burner behaviour in the very confined space affects the results. However,
equations (35) and (36) require knowledge of the real fire dimensions (at least the
fire width) as well as the heat release rate or fire intensity. A tunnel design cannot
be based on a particular fire geometry, but must consider a more general case, so
there is a need to consider nominal inputs that would lead to reasonably ‘safe’ val-
ues of critical velocity, without unnecessary over-prediction which also increases
risk and cost if used as a design basis.

6.1 Design assumptions

It is unlikely that a truck will become completely sideways in the tunnel. The
width of a truck is 2.5 m. Spilt liquid fuel will run to the gutter following the
cross-fall in combination with any longitudinal gradient component. It would start
as a splash zone and perhaps grow somewhat as it flows, depending on the road
surface slope. It is felt that 2.5 m would cover that also, noting that wider fires re-
duce the critical velocity.
The fire intensity can be taken as that of a hydrocarbon pool fire, adopting
2.25 MW/m2 as the figure. As shown by the Memorial Tunnel tests (see Table 4),
0.2 seems to be a reasonable value for ε (radiative heat fraction), and η (heat frac-
tion lost to water suppression) is taken as zero (no active suppression). Section
4.3 demonstrates that grade factor 𝐾𝑔 can also be taken as 1.0 for tunnels with typi-
cal gradients. The fire can conservatively be taken as starting at roadway level,
and so the buoyant convection scale 𝐿𝑛 can be taken as equal to the tunnel height
𝐻 . The empirical numbers can be adopted as listed in Table 2, as it was shown in
Section 5 that those values provide good approximation to real tests.
Making a sea level assumption, air density is 1.2 kg/m3with an ambient tempera-
ture of 294 K.

6.2 Outcome of design assumptions for typical tunnel profiles

To explore the variation of the critical velocity for typical two to three lane tun-
nels, the equations (35) and (36) were evaluated for different tunnel profiles with
the assumptions as noted in Section 6.1.
 39

Figure 19 and Figure 20 show the range of critical velocity values for a wide
variation of TBM and rectangular tunnel profiles. Table 10 lists the geometric pa-
rameters for the TBM tunnel profiles analysed.
When considering that the design HRR for road tunnels typically lies in the
range of 30 MW to 60 MW, the ‘design’ critical velocity in the TBM profiles var-
ies between 2.82 m/s and 3.63 m/s. For the same HRR range, the critical velocity
in the rectangular profiles varies between 2.47 m/s and 3.50 m/s. As discussed in
Section 4.4, the critical velocity values in rectangular tunnels are underestimated
by about 10%.

Table 10. Geometric parameters for typical TBM or arched tunnel profiles used for Figure 19.
Hydraulic diam- Tunnel
Pro- Area
Description eter height
file name (m2)
(m) (m)
3-lane arched tun-
TBM1 nel with false ceiling, 85.38 9.06 6.80
see Figure 14

TBM2 3-lane TBM tunnel 85.60 9.28 8.07

3-lane TBM tunnel,

TBM3 106.0 10.7 9.82
see Figure 12
2-lane arched tun-
TBM4 40.00 5.99 5.00
nel with false ceiling

TBM5 2-lane TBM tunnel 40.00 6.44 5.89

2-lane arched tun-

TBM6 59.95 8.13 7.86
nel, see Figure 8

Of course, in locking in the adoption of a 2.5 m wide fire and assuming that the
base of the fire is at the floor, the model results will no longer reproduce the Me-
morial Tunnel tests, for which the fire pans were elevated and 3.66 m wide.
40

Figure 19. Critical velocity values for different TBM tunnel profiles and total fire heat release
rates based on proposed design assumptions.

Figure 20. Critical velocity values for different rectangular tunnel profiles and total fire heat
release rates based on proposed design assumptions.

6.3 Simplified equation for design purposes

The design HRR for road tunnels is typically >10 MW, which allows the strong
HRR dependency of critical velocity at low HRR to be ignored, and therefore the
“min()” function in equation (36) to be deleted. The specific heat capacity can be
taken as 1.007 kJ/(kg·K) over the range of interest. With those assumptions and
 41

the substitutions according to Section 6.1 and Table 2, the critical velocity as
given by equations (35) and (36) only depends on the total HRR and some geo-
metric parameters (tunnel height, tunnel area and hydraulic diameter) as shown by
the simplified equations (37) and (38) below.
1
𝐻 𝐻3 ∆𝑇 2
(37)
𝑈𝑐 = (1.476 − 2.64 ) (𝑔 2 )
𝐴 𝐷ℎ 294 + ∆𝑇

−1
−8 𝑄̇
∆𝑇 = [2.92 ∙ 10−4 ∙ 𝑈𝑐 ∙ (7.14 + 𝐻) (0.95 + 𝑒 −8∙10 𝐻 )] (38)

In equations (37) and (38), the variables have the units noted in the nomenclature
section. Several parameters also have units, as tabulated below by their numerical
value.

Table 11. Units of parameters used in equations (37) and (38). The parameters resulted from a
reduction of (35) and (36) after substituting the noted design assumptions and the empirical pa-
rameters listed in Table 2. Therefore, the physical meaning of the parameters listed is related to
the substitutions and combinations of the different numbers.
Parame-
Unit Location
ter
1.476 - First factor in eq. (37)
2.64 m First factor in eq. (37)
294 (𝑇𝑎 ) °K Second factor in eq. (37)
- 2
2.92∙10 s/(m K) First factor in eq. (38)
4

7.14 m Eq. (38)

0.95 - Last factor in eq. (38)
-8
8∙10 m/W Exponent in exponential function of eq. (38)

The applicability of the simplified equations (37) and (38) is limited to the real
test data and simulations on which the validity of the empirical parameters listed
in Table 2 relies. However, with the broad agreement between the full equations
and a very wide range of experiments, and the fact that the simplification of the
equations involved conservative parameter choices, the simplified equations are
seen as useful for road tunnels.
42

7 Conclusions

A fresh approach to accommodating the physics of the zero-backlayering tunnel

fire flow field in a mathematical model has yielded equations with reasonable pre-
dictive capability for critical velocity. The model is based on non-dimensional
numbers characterizing the gas flow field in the tunnel with mixed convection.
The model was validated against CFD and experimental measurement data. The
work has produced the following outcomes, the first three of which are novel, with
the last being reconfirmed.

• Explained by geometry and fire intensity considerations, heat release

rate beyond a certain value (about 10 to 15 MW for realistic fires in real
road tunnels) causes no further significant increase in critical velocity,
matching the Memorial Tunnel results.
• Tunnel aspect ratio (width to height ratio) no longer requires an addi-
tional adjustment, being handled natively by the current formula.
• Corrections for typical tunnel grades are small enough to be ignored,
such that no grade correction factor is required in design.
• It is clear why scaling up from small models by using Froude number
(ignoring density, fire intensity etc.) leads to unreliable numbers.

Recommendations on design parameters are provided to reduce the complexity of

the proposed equation. Based on those substitutions, it seems that a value between
2.8 m/s and 3.8 m/s is sufficient to prevent backlayering in typical road tunnels for
typical design fires with heat release rates between 30 MW and 60 MW. Within
the limitations noted, the proposed equations are ready for application in tunnel
designs, with the cautions, as before, that critical velocity may not be so critical to
design, and that application to specific geometries and fire scenarios requires
thought.

Further comparison of the proposed model with any full-scale data and relevant
test data would be welcome. The determination of reasonable η values to account
for the effect of fixed firefighting systems on the critical velocity value may be the
subject of future work.

Statements and Declarations

Author Contributions: “Conceptualization, Michael Beyer and Conrad Stacey;

methodology, Michael Beyer and Conrad Stacey; validation, Michael Beyer; for-
mal analysis, Michael Beyer, Conrad Stacey and Günter Brenn; resources, Conrad
 43

Stacey; data curation, Michael Beyer; writing—original draft preparation, Michael

Beyer and Conrad Stacey; writing—review and editing, Michael Beyer, Conrad
Stacey and Günter Brenn ; visualization, Michael Beyer; supervision, Conrad
Stacey; project administration, Conrad Stacey; funding acquisition, Conrad
Stacey; investigation, Michael Beyer and Conrad Stacey. All authors have read
and agreed to the published version of the manuscript.” Please turn to the CRediT
taxonomy for the term explanation.
Funding: This research was funded primarily by Stacey Agnew Pty Ltd and
Delve Stacey Agnew LLC, through direct salary payments during the very consid-
erable work time spent by Michael Beyer and Conrad Stacey between December
2019 and January 2024 to develop the thoughts reported here.
Acknowledgments: We are grateful to Matthew Bilson and Joe Gonzalez for
passing on details of the Memorial Tunnel tests, including setup sketches and
9 CDs of the raw data to re-analyse. Thanks to Peter Sturm for support, and moti-
vation starting in late 2019 to start the journey into sorting out the issues with crit-
ical velocity formulae.
Conflicts of Interest: Dr Stacey is an owner of funding companies Stacey Ag-
new Pty Ltd and Delve Stacey Agnew LLC. As an author, Dr Stacey also had a
strong role in the design of the study and the modelling; in the collection, analysis,
and interpretation of data; in the writing of the manuscript, and in the decision to
publish the results. There are no other intersections between funding and author-
ship, and there are no conflicts.

References

1. Thomas, P.H. The Movement of Smoke in Horizontal Passages Against an Air

Flow. Fire Research Station 1968, Fire Research Note No 723.

2. Thomas, P.H. The Movement of Buoyant Fluid Against a Stream and the Venting
of Underground Fires. Fire Research Station 1958, Fire Research Note No 351.

3. De Ris, J. Duct Fires. Combustion Science and Technology Volume 2 – Issue 4,

239-258, 1970, https://doi.org/10.1080/00102207008952252.

4. Danziger, N. H.; Kennedy, W. D. Longitudinal Ventilation for the Glenwood

Canyon Tunnels. Fourth International Symposium on the Aerodynamics & Venti-
lation of Vehicle Tunnels, York, UK, 23-25 March 1982; BHRA Fluid Engineer-
ing (pp. 169-186).

5. Lee, C.K.; Chaiken, R.F.; Singer, J.M. Interaction between Duct Fires and Venti-
lation an Experimental Study. Combustion Science and Technology Volume 20,
59-72, 1979, https://doi.org/10.1080/00102207908946897.

6. Kennedy, W. D. Critical velocity: Past, Present and Future. Seminar of Smoke

and Critical Velocity in Tunnels, London, JFL Lowndes, 9-11 March 1996; (pp.
305-322).
44

7. Hinkley, P.L. The Flow of Hot Gases Along an Enclosed Shopping Mall – A
Tentative Theory. Fire Research Station. Fire Research Note No 807, 1970,
https://publications.iafss.org/publications/frn/807/-1/view/frn_807.pdf.

8. Heselden, A. Studies of fire and smoke behavior relevant to tunnels. 2nd Interna-
tional Symposium on Aerodynamics and Ventilation of Vehicle Tunnels, J1-1 to
J1-18, Cambridge, UK, 23-25 March, 1976.

9. Heselden, A.J.M. "Studies of fire and smoke behaviour relevant to tunnels",

Building Research Establishment, CP 66/78, Borehamwood, UK, 1978.

10. Guelzim, A.; Souil, J.M.; Vantelon, J.P.; Son, D.K.; Gabay, D.; Dallest, D. Mod-
elling of a Reverse Layer of Fire-Induced Smoke in a Tunnel. Proceedings of the
Fourth International Symposium on Fire Safety Science, 277-288, Ottawa, Can-
ada, 13-17 June, 1994.

11. Saito, N.; Yamada, T.; Sekizawa, A; Yanai, E.; Watanabe, Y.; Miyazaki, S. Ex-
perimental Study on Fire Behvior in a Wind Tunnel with a Reduced Scale Model.
Second International Conference on Safety in Road and Rail Tunnels, 303-310,
Granada, Spain, 3-6 April, 1995.

12. Lee, S. R.; Ryou, S. An Experimental Study of the Effect of the Aspect Ratio on
the Critical Velocity in Longitudinal Ventilation Tunnel Fires. Journal of Fire
Sciences 23, 119-138, 2005, https://doi.org/10.1177/0734904105044630.

13. Oka, Y.; Atkinson, G.T. Control of Smoke Flow in Tunnel Fires. Fire Safety
Journal 25, 305-322, 1995, https://doi.org/10.1016/0379-7112(96)00007-0.

14. Atkinson, G. T.; Wu, Y. Smoke Control in Sloped Tunnels. Fire Safety Journal
27, 335-341, 1996, https://doi.org/10.1016/S0379-7112(96)00061-6.

15. Wu, Y.; Bakar, M.Z.A. Control of smoke flow in tunnel fires using longitudinal
ventilation system – a study of the critical velocity. Fire Safety Journal 35, 363-
390, 2000, https://doi.org/10.1016/S0379-7112(00)00031-X.

16. Kunsch, J.P. Simple model for control of fire gases in a ventilated tunnel. Fire
Safety Journal 37, 67-81, 2002, https://doi.org/10.1016/S0379-7112(01)00020-0.

17. Hwang, C.C.; Edwards, J.C. The critical ventilation velocity in tunnel fires – a
computer simulation, Fire Safety Journal 40, 213-244, 2005,
https://doi.org/10.1016/j.firesaf.2004.11.001.

18. Li, Y.; Lei, B.; Ingason, H. Study of critical velocity and backlayering length in
longitudinally ventilated tunnel fires. Fire Safety Journal 45, 361-370, 2010,
https://doi.org/10.1016/j.firesaf.2010.07.003.

19. Li, Y.; Ingason, H. Effect of cross section on critical velocity in longitudinally
ventilated tunnel fires. Fire Safety Journal 91, 303-311, 2017,
https://doi.org/10.1016/j.firesaf.2017.03.069.

20. Li, Y.; Ingason, H. Corrigendum to “Effect of cross section on critical velocity in
longitudinally ventilated tunnel fires” [Fire Saf. J. 91 (2017) 303-311]. Fire
Safety Journal 110, 2019, https://doi.org/10.1016/j.firesaf.2019.102893.
 45

21. Li, Y.; Ingason, H. Discussions on critical velocity and critical Froude number for
smoke control in tunnels with longitudinal ventilation. Fire Safety Journal 99,
22-26, 2018, https://doi.org/10.1016/j.firesaf.2018.06.002.

22. Grant, G.B.; Jagger, S.F.; Lea, C.J. Fires in tunnels. Philosophical. Transactions
of the Royal. Society. London. 356, 2873-2906, 1998,
https://doi.org/10.1098/rsta.1998.0302.

23. Lönnermark, A. On the Characteristics of Fires in Tunnels. Doctoral Thesis, De-

gree of Doctor of Philisophy at Lund University, Sweden, 2005.

24. Ingason, H. state of the Art of Tunnel Fire Research. Proceedings of the Ninth In-
ternational Symposium. Fire Safety Science 9 33-48, 2008, DOI:
10.3801/IAFSS.FSS.9-33 https://publications.iafss.org/publica-
tions/fss/9/33/view/fss_9-33.pdf.

25. Agnew, N.; Tuckwell, B. Not So ‘Critical Velocity’, 16th Australasian Tunnelling
Conference; Sydney, Australia, 2017.

26. Haddad, R.K.; Maluk, C.; Reda, E.; Harun, Z. Critical Velocity and Backlayering
Conditions in Rail Tunnel Fires: State-of-the-Art Review. Hindawi Journal of
Combustion, 2019, Volume 2019, 20 pages,
https://doi.org/10.1155/2019/3510245.

27. MTFVTP, “Memorial Tunnel Fire Ventilation Test Program - Comprehensive

Test Report,” Bechtel/Parsons Brinckerhoff, Boston, 1995, http://www.tun-
nelfire.com/order.htm.

28. MTFVTP, “Memorial Tunnel Fire Ventilation Test Program - Memorial Tunnel
Test Data Report incl. 9 discs of raw data,” Bechtel/Parsons Brinckerhoff, Bos-
ton, 1995.

29. Kile, G.W.; Gonzalez, J.A. The Memorial Tunnel Fire Ventilation Test Program:
The Longitudinal and Natural Tests. ASHRAE Technical Data Bulletin 62-74,
Volume 13, Number 1, A Collection of Papers from the ASHRAE Meeting in
Boston, Massachusetts, June-July 1997.

30. Kile, G.W.; Gonzalez, J.A. The Memorial Tunnel Fire Ventilation Test Program:
The Longitudinal and Natural Tests. ASHRAE Transactions 103, ProQuest Sci-
ence Journals 701-713, 1997.

31. Ansys, Inc, “ANSYS Fluent User's Guide, Release Janury 2021 R1,” USA, 2021.

32. Ansys, Inc, “ANSYS Fluent Theory Guide, Release January 2021 R1,” USA,
2021.

33. Ansys, Inc, “ANSYS Fluid Dynamics Verification Manual, Release January 2021
R1,” USA, 2021c.

34. Attar, A.A.; Pourmahdian, M.; Anvaripour, B. Experimental Study and CFD
Simulation of Pool Fires. International Journal of Combustion Applications
(0975-8887), vol. Volume 70, no. No. 11, pp. 9-15, May 2013.

35. K. Karki, S. Patankar, E. Rosenbluth and S. Levy, CFD Modeler for Jet Fan Ven-
tilation Systems. BHR Group 10th ISAVVT, Boston, USA, 2000.
46

36. Stacey, C.; Beyer, M. Critical of Critical Velocity – An Industry Practitioner’s

Perspective, 10th International Conference ‘Tunnel Safety and Ventilation, Virtual
Conference Austria, 1st to 3rd of December 2020; Graz, Austria, 2020,
https://www.tunnel-graz.at/assets/files/tagungs-
baende/2020/13_Stacey_Beyer_Tunnel2020.pdf.

37. Stacey, C.; Beyer, M. Recorded presentation - Critical of Critical Velocity – An

Industry Practitioner’s Perspective, 10th International Conference ‘Tunnel Safety
and Ventilation, Virtual Conference Austria, 1st to 3rd of December 2020; Graz,
Austria, 2020, https://www.youtube.com/watch?v=-5rTvTnCm1M.

38. ASTRA 13001. (2008). Lüftung der Strassentunnel - Systemwahl,

Dimensionierung und Ausstattung (2008 V2.03). Bern, CH: Richtlinie, Bun-
desamt für Strassen ASTRA.

39. RABT. (2016). Richtlinie für die Ausstattung und den Betrieb von Straßentunnel
(RABT). Germany: Richtlinie, Forschungsgesellschaft für Straßen- und
Verkehrswesen , Arbeitsgruppe Verkehrsführung und Verkehrssicherheit.

40. RVS 09.02.31. (2014). Tunnel / Tunnel Equipment / Ventilation Systems - Basic
Principles. Wien, AT: Richtlinie, Österreichische Forschungsgesellschaft Straße -
Schiene - Verkehr.

41. PIARC (C3.3). (2011). Operational Strategies for Emergancy Ventilation. (P. C.
(C3.3), Ed.) World Road Association (PIARC).

42. PIARC (C5). (1999). Fire and Smoke Control in Road Tunnels. (P. C. (C5), Ed.)
World Road Association (PIARC).

43. Sturm, P.; Beyer, M.; Rafiei, M. On the problem of ventilation control in case of
a tunnel fire event. Case Studies in Fire Safety, CSFS 22, Elsevier publishing,
2015 https://doi.org/10.1016/j.csfs.2015.11.001.

44. NFPA 502. (2014). Standard for Road Tunnels, Bridges, and Other Limited
Access Highways. NFPA 502, 2014.

45. NFPA 502. (2017). Standard for Road Tunnels, Bridges, and Other Limited
Access Highways. NFPA 502, 2017.

46. NFPA 502. (2020). Standard for Road Tunnels, Bridges, and Other Limited
Access Highways. NFPA 502, 2020.

47. NFPA 502. (2020). Tentative Interim Amendment (TIA), TAI 20-1, Reference:
Annex D, 11.3, A.11.4.2, Annex O, 26 th August 2021, https://www.nfpa.org/as-
sets/files/AboutTheCodes/502/TIA_502_20_1.pdf.

48. NFPA 502. (2023). Standard for Road Tunnels, Bridges, and Other Limited
Access Highways. NFPA 502, 2023.

49. Beyer, M.; Stacey, C.; Dix, A. Critical velocity and the significance of the immi-
nent retraction of 2020 NFPA 502’s Annex D critical velocity equations Part
One. Australian Tunnelling Society, 2021,
https://www.ats.org.au/2021/03/11/critical-velocity-a-cautionary-note-to-practi-
tioners/.
 47

50. Beyer, M.; Stacey, C.; Dix, A. Critical velocity and tunnel smoke control Part 2,
Filling the NFPA 502 void. Australian Tunnelling Society, p. 6, 2021,
https://www.ats.org.au/2021/05/03/critical-velocity-and-tunnel-smoke-control-
part-two/.

51. Beyer, M.; Stacey, C. CFD Validation for Tunnel Smoke Control Design. In
ITnA-Reports Volume 105, 11th International Conference Tunnel Safety and
Ventilation, Graz, Austria, 9th and 10th of May 2022; DOI: 10.3217/978-3-85125-
883-7-09; https://openlib.tugraz.at/download.php?id=6315edb636dbb&loca-
tion=browse.

52. Shi, Y. S.; De Los Rios, N.; Lopez, K. The Critical Penalty. 19th International
Symposium on Aerodynamics, Ventilation & Fire in Tunnels, Brighton, UK, 28th
to 30th of September 2022.

53. Shi, Y. S.; Bowman, I.; De Los Rios, N.; Harvey, N.; Lopez, K.; Pelessone, L.;
NFPA 502 Critical Velocity Calculation Methodologies and Dimensionless Heat
Release Rate. 10th International Symposium on Tunnel Safety and Security, Sta-
vanger, 513-525, Norway, 26th to 28th of April 2023.

54. Karlsson, B.; Quintiere J. G.; Enclosure Fire Dynamics. CRC Press Boca Raton
London New York Washington, D.C., 2000, ISBN 0-8493-1300-7.

55. Rhodes, N.; CFD Predictions of Memorial Tunnel Fire Test No. 614,” 2nd Inter-
national Conference on Safety in Road and Rail Tunnels, Granada, organized by
University of Dundee and ITC Ltd., 1995.

56. Galdo Vega, M.; Argüelles Díaz, K.M.; Fernández Oro, J.M.; Ballesteros Taja-
dura, R.; Santolaria Morros, C.; Numerical 3D simulation of a longitudinal venti-
lation system: Memorial Tunnel case. Tunnelling and Underground Space Tech-
nology Volume 23, Issue 5, 539-551, September 2008,
https://doi.org/10.1016/j.tust.2007.10.001.