Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

International Journal of Thermal Sciences 200 (2024) 108957

Contents lists available at ScienceDirect

International Journal of Thermal Sciences


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

Experimental study on the smoke diffusion length affected by canyon winds


in the open section of railway tunnel groups
Xin Chen , Li Yu , Jianxun Huo *, Xue Wang , Songshen Wang , Xiaohan Guo , Mingnian Wang
Key Laboratory of Transportation Tunnel Engineering, Ministry of Education, Southwest Jiaotong University, Chengdu, 610031, China

A R T I C L E I N F O A B S T R A C T

Keywords: There are numerous tunnel groups, referring to multiple tunnels spaced a certain length apart. When a fire
Railway tunnel group occurred in a railway tunnel group, the train should brake out of tunnel portal to evacuate passengers. However,
Canyon wind tunnel groups are usually located in canyon areas, where the canyon winds are common, and the canyon wind
Yaw angle
would have a significant effect on the diffusion of fire smoke in the open section of the tunnel group.
Smoke diffusion length
In order to better grasp the diffusion characteristics of fire smoke in the open section affected by canyon wind,
a series of 1/17 modeling experiments were conducted, a full-field, instantaneous and non-invasive measurement
technology (PIV) was applied.
The results showed that (1) The main factors in canyon winds that affected smoke diffusion were wind velocity
and yaw angle. (2) As canyon wind velocity increased, the smoke diffusion length in the open section decreased.
(3) As the yaw angle increased, the smoke diffusion length in the open section increased. (4) A prediction method
of smoke diffusion length affected by canyon winds is established, which error is not more than 10%. It could
determine whether the fire smoke will endanger passengers’ evacuation in case of natural diffusion, and provide
a basis for setting up smoke control ventilation.

1. Introduction should stop at the open section. However, when the length of the open
section is too short, fire smoke may diffuse into the tunnel, which will
Internationally, a group of tunnels with a spacing of less than a endanger the evacuation safety seriously. There are no provisions in the
certain length is generally considered to be a tunnel group. For exam­ UIC and SIA on fire prevention in the open sections of tunnel groups. TSI
ples, in China, the spacing is the length of a train [1], and in Japan, it is defines tunnel groups that when the separation between tunnels in open
500 m. In the event of fire, the general strategy is to get the train out of section is longer than the maximum length of the train intended to be
the tunnel in order to reduce the consequences of the fire if possible. It is operated on the line +100 m, two or more consecutive tunnels will be
widely recognized in many international organizations and countries, considered as a tunnel group. It means that TSI recognizes that fire
such as China [1], UIC [2], TSI [3], Switzerland [4]. Therefore, the smoke is not a hazard to the tunnel environment at distance greater than
location of the fire source is in the open section of the tunnel group. The its mentioned above. Maybe it’s because of the small number of tunnel
diffusion of fire smoke is mainly affected by the fire power, the wind groups in European countries, the safety distance they suggest is a little
field, and the form of the structure. The wind field includes the nature long. Similarly, the minimum tunnel distance in Chinese tunnel groups is
wind inside the tunnel and the canyon wind in the open section. Canyon also defined based on the train length. On this basis, however, China has
winds are a common phenomenon in mountainous areas. There are further regulations. The China’s code (TB10020-2017) [1] gives a crit­
three main conditions for the formation of canyon winds [5], as follows: ical smoke diffusion length of the open section, which is 250 m. When
(1) Canyon terrain, which can restrict and accelerate airflow, which the length of the open section is less than 250 m, ventilation is required
contribute to the formation of canyon winds. (2) Stable atmospheric, at the tunnel portal to control the smoke outside the tunnel. Li (2017)
which ensures that canyon winds are less disturbed. (3) Difference in [6] conducted modeling experiments with a scaling ratio of 1/13 and
temperature, which is an important driver for canyon wind. studied the critical smoke diffusion length in the open section affected
If a fire breaks out while the train is in a tunnel group, the train by natural wind inside tunnels. The results showed that at the maximum

* Corresponding author.
E-mail address: cx@my.swjtu.edu.cn (J. Huo).

https://doi.org/10.1016/j.ijthermalsci.2024.108957
Received 15 November 2023; Received in revised form 17 January 2024; Accepted 5 February 2024
Available online 22 February 2024
1290-0729/© 2024 Published by Elsevier Masson SAS.
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Fig. 1. The velocity distributions and three-dimensional direction of canyon wind.

natural wind speed, the critical diffusion length was 50 m. Zhao et al. effect of canyon wind velocity on the smoke movement in tunnel. The
(2021) [7] studied the diffusion characteristics of fire smoke from results showed that the airflow in the tunnel caused by the canyon wind
adjacent tunnels under different nature wind velocities and developed a represented different characteristics at different positions, the airflow in
prediction model for the critical crossflow velocity considering different the tunnel could be divided into three areas named as negative pressure
open section lengths. Wang (2021) [8] simulated the braking process of zone, transition zone, and unidirectional flow zone. Normally, the
a fire train in the open section of a tunnel group by CFD, and investigated smoke diffusion length at the side of tunnel portal decreased with the
the effect of piston wind on smoke diffusion. increasing canyon wind velocity, except for the fire located in the
In past studies of tunnel group fire, people usually ignored the effects negative pressure zone, no matter how large the canyon wind was, the
of canyon winds. Nevertheless, due to the advantages of tunnels for fire smoke would overflow from the downstream tunnel portal. Luan
transportation in canyon areas, a large number of tunnel groups have et al. (2021) [17] and Shen et al. (2021) [18] experimentally studied the
emerged. In actual engineering, canyon winds are common and have a effects of canyon wind velocity and yaw angle on the smoke
significant impact on fire smoke diffusion, where tunnel-bridge-tunnel is back-layering length, respectively. The prediction equations for the
the most common structural form. In terms of wind resistance of bridges, back-layering length under different wind velocitied and yaw angles
a great deal of research had been carried out and relatively mature codes were established. The smoke back-layering length decreased with the
had been developed, such as China [9], European countries [10]. They increasing canyon wind velocity until the back-layering length
usually consider wind as wind loads, and usually focus on extreme cases decreased to 0, and the smoke back-layering length increased with the
of high winds. For example, Chinese code adopts the 10-min average increasing angle when canyon wind deflected into the tunnel. Ao et al.
wind velocity with a return period of 100 years as the design reference (2021) [19] also conducted experimental research, mainly focused on
wind velocity. Whereas a fire is a happenstance event, smoke diffusion fire HRR and flame characteristics. A normalized mass burning rate
should consider the winds, which are common and periodic. Thus, the model considering the distance to the upstream portal and the yaw
conclusions of wind-resistant design of bridges cannot be directly angle, and a prediction correlation between the flame tilt angle and the
adopted. Some scholars classified winds according to their characteris­ canyon cross wind speed and yaw angle were developed. The study
tics, which was of great instructive significance. In meteorology, indicated that (1)when the canyon wind deflected into the tunnel portal,
mountainwind systems were classified into large-scale synoptic and the mass burning rate and the flame tilt angle decreased with the
local-scale windsystems [5]. Moreover, local-scale windsystems could increasing wind yaw angle, (2) when the canyon wind deflected out of
be classified into Sudden intense wind, Cooling windstorm [11], The the tunnel portal, the mass burning rate and the flame tilt angle
periodic thermally-driven wind (PTDW events) [12]. PTDW events were remained steady and independent of yaw angle. However, in the above
the most common wind in canyon areas [13], which should be paid more studies, the fires located inside the tunnel did’t conform to the general
attention in the study of fire smoke diffusion. Studies showed that daily principle for the fire in railway tunnel group. And the studies focused on
wind in the canyon was cyclical and periodic. It was higher in the middle the indirect effects of canyon winds on fires inside the tunnel.
of the day and lower at night, and higher in the winter and lower in the In this paper, in order to grasp the diffusion characteristics of fire
summer [14]. Due to the constraints of the mountains on either side of smoke in the open section affected by canyon wind, modeling experi­
the canyon, the direction of canyon wind was ususlly locked in Ref. [15]. ments with a scaling ratio of 1/17 were conducted. Particle image ve­
The above studies on wind can provide a basis for the study of the locity measurement technology (PIV), an advanced flow field
diffusion law of fire smoke. measurement technology with advantages of full-field, instantaneous
Some scholars also studied the effect of canyon winds on tunnel and non-invasive, was applied. The effects of canyon wind velocity and
group fire, mainly focused on the situation where fire source was in the yaw angle on smoke diffusion length in open section were investigated.
tunnel 0~330 m from the portal. Zhao et al. (2021) [16] studied the Finally, combined with numerical simulation results, a prediction model

2
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

their values through literature survey. Section 3 introduced the estab­


lishment of modeling experiments and analyzed the effects of canyon
wind velocity and yaw angle on smoke diffusion length. Section 4 car­
ried out numerical simulations and verified them with experimental
results, a prediction model for fire smoke diffusion length was developed
and the errors between the predicted results and experimental results
were analyzed.

2. Statistical characteristics of canyon winds

Fig. 2. The classification of canyon wind. 2.1. The characteristics of canyon wind

When the airflow passed through a canyon, its cross section de­
creases and the airflow was accelerated due to the "Venturi Effect” to
form the canyon wind [20]. Canyon wind was not a constant airflow. It
was characterized by velocity distributions and three-dimensional di­
rection in the space, as shown in Fig. 1. In the vertical direction, wind
velocity increased as the height above canyon bottom increased and
stopped increasing when the height reached the height of gradient wind.
Chinese code [9] assume that the variation follows an exponential law.
In the project, according to the weather station, the basic velocity was
determined, and then the design velocities at different height could be
calculated, as shown in Fig. 1(b). Numerous studies [21,22] indicated
that wind profiles along the bridge was showed in Fig. 1(c). The further
away from the mountain, the friction got decreased and the wind ve­
Fig. 3. The area of canyon wind affecting the smoke diffusion in the locity got increased. At a certain distance, there was very little frictional
open section. effect, however, the compression of the canyon diminished and the wind
velocity decreased slightly in the middle of the bridge. In addition to
velocity, direction was also an important characteristic of canyon winds.
Table 1 Due to the constraints of the mountains on either side of the canyon, the
Survey of wind yaw angles in typical bridge engineering. canyon winds are usually locked in Ref. [15]. However, the wind will be
Location Direction of Direction of Yaw Scholar deflected locally due to the obstacles. Chinese code [9] describes the
the bridge main average angle direction of the wind in the horizontal and vertical planes as the yaw
axis (◦ ) wind speed (◦ ) (◦ ) angle and the attack angle, as shown in Fig. 1(d). The attack angle is the
A bridge located in 220 290 20 Yu et al. angle between the wind direction and the horizontal plane. The yaw
typical V-shaped (2019) [20] angle is the angle between the projection of the wind direction in the
gorge
horizontal plane and the vertical plane of the bridge axis.
Xingkang Bridge in 120 152~164 16~28 Zhang et al.
the southwestern (2022) [28]
According to the scale and range of airflow, the wind were classified
mountainous into large-scale synoptic and local-scale wind systems [5]. The former
region of China was caused by the interaction between large-scale atmospheric circu­
The Hardanger 155 75 10 Fenerci lation and topography [23], such as the cold front when macro-scale
Bridge in et al. (2017)
cold air passed through mountainous terrain. The latter was formed
Hordaland county [29]
of Norway over the local area in mountainous landscapes, such as the periodic
Jin’an Bridge in 100 172 18 Li et al. thermally-driven winds (PTDW events) [12]. Moreover, sudden intense
Jinsha River, China (2022) [30] winds or thunderstorms driven by strong local convection in the region
The Sulafjord Bridge 80 150、180 10、20 Castellon
also belonged to local-scale wind systems [24], which all had evident
in a Norwegian et al. (2022)
fjord [31]
local terrain dependence. Thus the winds could be categorized as shown
A bridge in the 257 314~323 24~33 Yang et al. in Fig. 2. Among them, PTDW events were the most common wind in
mountains of (2022) [32] canyon areas, which were developed by the local temperature gradient.
Chongqing, China Thus canyon winds varierd in cycles of days, with high velocity during
The Hardanger 160 45~80 0~25 Lystad et al.
the day and low velocity at night [25]. Also, there were different vari­
Bridge in Norway (2018) [21]
Anqing Yangtze River 0 33 57 Yang et al. ation in different seasons [26].
Bridge (2018) [33]
the Xihoumen 40 160 30 Ma et al. 2.2. Major wind characteristics affecting smoke diffusion
suspension bridge (2020) [34]
in Xihoumen Bay,
China The purpose of the studying smoke diffusion in the open section was
Lingdingyang bridge 90 157 23 Liu et al. to determine whether the smoke would enter the tunnels and endanger
in the Pearl River, (2022) [35] the safety of evacuation. The fire smoke would maintain a unidirectional
China flow under the influence of thermal buoyancy and wind. If the smoke
flowed upwards out of the tunnel vault height or laterally out of the side
for fire smoke diffusion length in open section under the influence of walls, the smoke couldn’t return, which was harmless to evacuation. The
canyon wind was established, which could correct the critical diffusion canyon wind area studied in this paper, which affected the smoke
length of fire smoke in the open section in the code. diffusion in the open section, was the limited space above the bridge to
The structure of this paper was as follows. Section 2 sifted the major the tunnel vault height, as shown in Fig. 3.
characteristics of canyon winds affecting smoke diffusion and identified The fire smoke diffusion length was the key to the study, which was
the result of thermal buoyancy competing with wind. The magnitude of

3
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Table 2 2.3. The values of canyon wind in the study


A list of scaling correlations for the tunnel group model.
Parameter Model scale Equation number As mentioned in section 2.2, the wind could be classified into three
5/2 categories, including PTDW events, sudden intense winds, and large-
Heat release rate (kW) Qm/Qf=(lm/lf) (5)
Velocity (m/s) vm/vf =(lm/lf)1/2 (6) scale cold fronts, in which PTDW events was the most common wind
Temperature(K) Tm/Ts = 1:1 (7) and periodic in mountainous areas. The other two were contingent,
Time(s) tm/tf=(lm/lf)1/2 (8) more disturbed by weather, and significantly larger than PTDW events.
Volume (m3/s) Vm/Vf=(lm/lf)5/2 (9) Due to the extremely low probability of tunnel fires, the rare canyon
winds were not considered. For the sudden intense wind and the cooling
thermal buoyancy was determined by the HRR of the fire source and the windstorm, recent studies have given the screening criteria [11,27]. One
direction was always vertically upward. In terms of canyon winds, the point was in common, that maximum instantaneous wind velocity was
corresponding factors were wind velocity and direction. As mentioned in greater than 15 m/s. Therefore the considered values of wind velocities
section 2.2, wind direction included the attack angle and the yaw angle. in the study were the 0–15 m/s.
The range of the attack angle considered in the actual project was The canyon wind yaw angle was determined by the direction of the
− 3◦ –3◦ [9]. Even if the canyon wind reached 15 m/s, the vertical bridge axis and the canyon terrain, which varieed from project to
component of the canyon wind at an attack angle of 3◦ was only 0.8 m/s, project. The yaw angles based on field measurement of 10 typical
which was very small compared to thermal buoyancy. Numerical bridges were surveyed, as shown in Table 1. The results showed that
simulation results also showed that this range of the attack angles had most of the yaw angles were less than 30◦ , and the maximum yaw angle
little effect on the smoke diffusion. Thus, major wind characteristics was 57◦ . Therefore the considered values of yaw angles in the study were
affecting smoke diffusion were wind velocity and yaw angle. The the 0–60◦ .
following studies were conducted based on the above two
characteristics.

Fig. 4. Experimental model.

4
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Fig. 5. The tunnel cross-section size.

coefficient of air was 1.9 × 10− 5 m2/s. The Reynolds number of the
prototype was greater than 105, as shown in Eq (3). And the Reynolds
number of the model could be expressed as Eq (4a). Therefore, the ex­
periments were conducted in a 1/17 scale, according to the scale
requirement as shown in Eq (4b). And the experimental device consisted
of a tunnel group and a series of wind turbine components, as shown in
Fig. 4.

up dp 1.9 × (20000 × 0.7)1/5 × 11.21


= = 75.65 × 105 > 105 (3)
ν 1.9 × 10− 5
( )1/2 ( ) ( )3/2
up dm
dp
dp ddmp 1.9 × (20000 × 0.7)1/5 × 11.21 × dm
dp
um dm
= = > 105
Fig. 6. The rectangular rectifier. ν ν 1.9 × 10− 5

(4a)
3. The effect of canyon winds on smoke diffusion
dm 1
> (4b)
3.1. Experimental setup dp 17.89

The 1/17 scale model tunnel group consisted of a 3 m tunnel and a 6


Model experiment was a common method used in tunnel fire studies. m tunnel with the longitudinal slope of 30‰. The distance between
The froude model was used to establish similarity ratios between the these two tunnels caould be adjusted between 0.58 m–1.8 m, as shown
scaled-down and full-size models, and the results of the study were in Fig. 5(a). The tunnel cross-section was simplified to straight edge wall
applied to actual fire scenarios. The dimensional relationship of each type with a height of 0.58 m and a width of 0.82 m. The elevation slope
parameter in the fluid was obtained by Williams (1969) [36]. Scaling of the tunnel entrance side was 60◦ . Two baffle plates, with a height of
correlations of some key parameters were presented in Table 2. In 1.0 m and a width of 1.5 m, were set at the entrances to simulate the
addition, Froude analogy requires that the Reynold number (Re) of the mountain of the tunnel entrance, as shown in Fig. 5(b). The framework
fluid flow in the model and prototype must be in the turbulent and vault of the tunnel were made of 2 mm thick iron plates. In order to
self-simulation region, where the flows can be regarded as analogous. In observe the flow of smoke in the tunnel, the two side walls and the
general, this means the Re should be greater than 105, and it can be bottom panel of the tunnel were made of 1.5 mm thick acrylic.
shown as Eq (1) and Eq (2). Two axial fans (Fans A) were placed in the open section of the tunnel
up dp group to simulate the canyon wind. In order to create a more realistic
> 105 (1) canyon wind environment, the outlet of fans A should completely cover
v
the open line section of the tunnel group. A 1 m × 1.8 m × 2 m (height ×
um dm width × length) rectangular rectifier was fitted at the outlet of fans A.
> 105 (2)
v Besides, 3 layers of grids were placed in the rectifier to homogenize the
turbulence at the outlet of fans A. The diameter and spacing of griller
where:up and um are the characteristic velocities of prototype and model, round holes were 10 mm and 15 mm respectively. The rectifier was
m/s. dp and dm are the equivalent diameters of prototype and model, m. ν shown in Fig. 6. The airflow velocity of Fans A could be adjusted from
is the viscosity coefficient of fluid motion, m2/s. 0 to 15 m/s by adjusting the frequency from 0 to 50 Hz. And different
Referring to the relevant codes, the heat release rate of a train fire wind yaw angles could be set by turning fans A and the rectifier. The
was 20 MW. Considering larger cross section could make the smoke other two axial fans (Fans B) were placed at both portals of the tunnel
diffuse into the tunnel easilier. Therefore, the railway tunnel section in group to simulate longitudinal natural winds in the tunnel. Fans B were
this experiment was decided to be a two-lane. And the viscosity

5
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Fig. 7. Arrangement of temperature measurement points.

Fig. 8. PIV measurement system.

(pulse energy ≥200 MJ, wavelength: 532 nm, pulse width: 6–10ns), a
synchronization controller (timing error <0.25ns). Instead of using
externally dispersed tracer particles, in this experiment, fire soot parti­
cles were used. PIV could measure the flow of fire smoke directly. Placed
the PIV at tunnel portal to capture smoke motion characteristics, as
shown in Fig. 8.
Thermal velocimeter, with the resolution of 0.001 m/s, was adopted
to measure the wind velocity, as shown in Fig. 9. The relevant velocities
in the experiment were calibrated by multipoint averaging.
In the experiment, burning gasoline was used as the fire source.
Considering a heat release rate of 20 MW for a real train fire, there was a
heat release rate of 16.8 kW for the oil pool fire in the experimental
model. Numerous studies of heat release rates at different gasoline sump
pan areas have been carried out at USTC [45]. According to the study,
Fig. 9. Thermal velocimeter. the size of the oil pool was designed as 12.3cm × 12.3cm × 5 cm. 150 ml
of gasoline was burned in each case.
also controlled by two frequency converters and their air velocities
could be adjusted from 0 to 3.5 m/s. 3.2. Experimental cases
To measure smoke temperature, there were 11 groups of K-type
thermocouple beams in the center line of the two tunnels. The mea­ There were three assumptions in this paper:
surement error of K-type thermocouple was 0.1 ◦ C. A group of K-type
thermocouple beam included 5 measurement points, whose distance (1) Assuming that the canyon wind is a uniform wind flow, the wind
from the tunnel vault was 10 mm, 115 mm, 230 mm, 345 mm and 460 velocity and yaw angle in the canyon remain constant. This has
mm respectively, as shown in Fig. 7. been common in past studies [16,18,46].
In order to measure the flow field of the tunnel group, particle image (2) Assuming that the train fire area is small, an oil pool fire is used as
velocity measurement technology (PIV) was adopted. PIV can obtain a substitute [47]. It will be used as a basis to study the longitu­
transient velocity field by digital image processing and mathematical dinal effect length of fire smoke in open section. The relevant
algorithms. There was no need to contact the measured flow field during code also divides the length of the fire’s effect into inside-train
the whole measurement process with the advantages of full-field, and outside-train components. This assumption was adopted in
instantaneous and non-invasive [37]. It have been widely used for this paper.
flow field testing of water currents [38,39], air currents [40,41], su­ (3) Assuming that the fire train always stops in the center of the open
personic velocities [42,43] and other types [44] of flow fields. section, accordingly, the oil pool is placed in the middle of the
The PIV system in this experiment was manufactured by MicroVec in open section [6]. It also means that the distance of the fire source
China. It consisted of a software analysis part and a CCD image acqui­ from the portal of the tunnel is one-half the length of the open
sition and processing hardware part, including a Dual-exposure cross- section.
frame CCD camera (resolution >16 megapixels), a dual-pulse laser

6
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Table 3 diffusion lengths and yaw angles. Based on the results of the pre-
Experimental cases. experiments and the requirement for comparison with numerical sim­
No. Open section Length Canyon Wind yaw Canyon wind velocity v ulations, the experimental cases were determined. Table 3 summarizes
D (m) angle β (◦ ) (m/s) all the experimental variables for this experiment, including 5 open
1–3 10 0 4,5,6 section lengths, 5 wind yaw angles and several canyon wind velocities.
4–7 30 2,7,8,9 There were 51 experimental cases implemented. In addition, the longi­
8–10 60 8,9,10 tudinal slope of the tunnel group was 30‰ and the slope angle at the
11–13 15 0 2,3,4 tunnel portal was 60◦ . Considering the most unfavorable case, the lon­
17–19 30 5,6,7
20–22 20 0 1,2,3
gitudinal wind direction in tunnels was along the uphill direction. And
23–26 30 0,2,4,5,6 its velocity was 2.5 m/s [48]. In the experimental cases, the varying of
27–30 60 2,8,9,10 the fuel’s mass was recorded. The five-point difference method recom­
31–33 25 0 0,1,2 mended in ISO 5660-1: 2015 was used to calculate the mass loss rate.
34–36 30 2,3,4
And HRR can be obtained as follows:
37–39 30 0 0,1,2
40–42 15 0,1,2
Q = ηm′ΔHc (10)
43–45 30 1,2,3
46–47 45 3,4,5
49–51 60 8,9,10 where: Q is the heat release rate, W. η is combustion efficiency, it’s 0.75
referring to Hu et al. (2009) [49]. m’ is mass loss rate of fuel, kg/s. ΔHc is
calorific value of fuel, and the gasoline’s is 45000 kJ/kg. Hu’ study
indicaded that HRR is different under different wind velocity and in­
creases with the increase of wind velocity. The HRR in this study is
consistent with Hu’s conclusion, The curves of HRR for different canyon
wind velocities were shown in Fig. 10. Took wind velocity 0 m/s as an
example, it was clear that HRR curves can be divided into the growth
stage, steady stage, and decay stage. The steady stage was the most
dangerous stage of a fire due to its maximum HRR and longest lasting
time. Therefore, HRR during the steady stage had been the subject of
numerous studies [50–54]. This paper also focused on the diffusion of
fire smoke during the steady stage and the average HRR during the
steady stage is assumed as the constant HRR.
When smoke did not enter the tunnel, PIV provided a relatively
complete measurement of smoke spread length and distribution. When
the smoke entered the tunnel, the PIV couldn’t measure the complete
smoke distribution, but it could be reacted by the temperature mea­
surement points in the tunnel. Based on the above assumptions and
measurements, an experimental procedure was developed as shown in
Fig. 10. HRR for different canyon wind velocities.
Fig. 11. In the experiments, each case was repeated twice to avoid
serendipity.

3.3. Effect of wind velocity on the fire smoke diffusion

To study the effect of wind velocity alone, the yaw angle and the
open section length were kept constant at 30◦ and 20 m, only the wind
velocity was varied. The distribution and flow lines of the fire smoke at
different canyon wind velocities were shown in Fig. 12. The diffusion of
fire smoke in the open section was dominated by thermal buoyancy, and
the direction of diffusion was predominantly vertical. However, in the
near area of the tunnel portal, the fire smoke was affected by the suction
wind from the portal, and the direction of smoke diffusion was deflected.
At all three wind velocities, there were similar phenomenons and smoke
all entering the tunnel but with different thickness of smoke layer. When
the canyon wind velocity was 0, it was a blank condition, indicating no
canyon wind.
Since the fire source in this study was in an open space, heat was not
easily gathered. In addition, the wind at the tunnel portal led to a large
amount of heat loss, and temperature might not be the main factor that
endangered the evacuation. However, it was the sinked soot particles,
which heat was lost. In the experiments, the soot particles could be seen
due to the laser, so the thickness of the smoke layer at the tunnel portal
Fig. 11. Experimental procedure. was measured based on the distribution of the particles. The grayscale of
the soot mocrogroup and background were uniform yet different. Thus
the grayscale histogram had a bimodal characteristic. We set the valley
Before formal experiments, some pre-experiments were conducted.
in the histogram to the threshold value and made soot mocrogroups
The purpose of the pre-experiments was to get the effect of canyon wind
distinguish from the background, which could be identified easily [55].
velocity and yaw angle on the fire smoke diffusion and to get the rough
The distance of the lowermost soot microgroup from the tunnel vault
range of critical canyon wind velocity corresponding to different smoke
was considered to be the thickness of the smoke layer. Based on a

7
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Fig. 12. The distribution and flow lines of the smoke at different wind velocities
Note: The shooting surface was at the centerline of the tunnel when 0 m/s, and near the side wall when 2 m/s, 4 m/s.

pixel-to-length conversion relationship, the thickness was measured. At temperature errors were shown in Table 4. Finally, the average values
the wind velocity of 0 m/s, the thickness was 6.36 m (374mm × 17). were used for analysis. At the wind velocity of 0, ΔTmax inside the tunnel
When the wind velocities was 2 and 4 m/s, the thickness was 4.73 m was 12.01 ◦ C. When the wind velocity was 2 and 4 m/s, ΔTmax inside the
(278mm × 17) and 4.42 m (260mm × 17) respectively, which were tunnel was 8.62 ◦ C and 2.42 ◦ C respectively, which was reduced by
reduced by 1.63 m and 1.94 m. This indicated that wind velocity was a 3.39 ◦ C and 9.59 ◦ C. As canyon wind velocities increased, the temper­
favorable effect. As the canyon wind velocity increased, the thickness of ature inside the tunnel decreased. This was attributed to two reasons: (1)
the smoke layer decreased, which effect was noticeable at lower canyon the increased canyon wind speed led to a decrease in smoke intake into
wind velocities. the tunnel, and (2) the increased canyon wind speed caused strong
The flow fields at different wind velocities were analyzed using the convection which reduced the temperature of the smoke itself. In
commercial software Micro Vec v3.6.5 and they were shown in Fig. 13. addition, it could be observed that the temperature of the smoke started
Because the tracer particles in the study were fire soot and there were no to decrease after 20 m into the tunnel, and it indicated that the smoke
additional tracer particles, the calculated flow fields didn’t completely started to cool and sink.
fill the filmed areas. It can be seen that the flow fields at different canyon In conclusion, as canyon wind velocity increased, the thickness of the
wind velocities at the tunnel portal were basically equal to the natural smoke layer and the temperature inside the tunnel decreased. When the
√̅̅̅̅̅̅
wind velocity inside the tunnel which is 0.5 m/s (2.5/ 17 ≈0.6 m/s). In wind speed increased by 2 m/s, the thickness of the smoke layer
general, due to the long air column inside the tunnel, canyon wind ve­ decreased by 0.3 m–1.63 m and the temperature decreased by
locity wouldn’t significantly affect the original flow field in the tunnel. 3.39 ◦ C–6.2 ◦ C.
Outside the tunnel, soot velocitiy decreased with increased canyon wind
velocitiy. When v = 0 m/s, the maximum soot velocitiy was 1.5 m/s. 3.4. Effect of wind yaw angle on the fire smoke diffusion
When v = 2 m/s and v = 4 m/s, the maximum soot velocities were 1.2
m/s and 0.8 m/s. This velocity was mainly driven by thermal buoyancy. To study the effect of wind yaw angle alone, the wind velocity and
Increased canyon wind velocitiy led to increased flow intensity, which the open section length were kept constant at 2 m/s and 20 m, only the
reduced soot temperatures. That was why soot velocitiy decreased. yaw angle was varied. The distribution and flow lines of the fire smoke
And the variation of temperature was used to reflect the character­ at different yaw angles are shown in Fig. 15. Smoke entered the tunnel at
istics of smoke, which had been widely acknowledged [56]. The effect of the yaw angles of 30◦ and 60◦ , not at the yaw angles of 0◦ .When the yaw
canyon wind velocity was further investigated below by the temperature angles was 0◦ , the smoke likewise diffused to the tunnel portal. Some­
in the tunnel. In order to eliminate differences in ambient temperature thing different was that the smoke was still dominated by thermal
during the experiments, all temperature values were replaced by tem­ buoyancy and the direction of diffusion was not deflected. When the yaw
perature rise. The temperature rise of the tunnel vault at different wind angle was 30◦ and 60◦ , the thickness of the smoke layer was 4.73 m
velocities was shown in Fig. 14. Each case was repeated twice and the (6.36mm × 17) and 9.33 m (549mm × 17) respectively. This indicated

8
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Fig. 13. Flow field at different wind velocities.

Table 4
The temperature errors at different wind velocities.
Canyon wind velocity (m/s) Absolute error (◦ C) Relative error (%)

Maximum Mean Maximum Mean

0 0.48 0.13 4.6 1.3


2 0.28 0.12 5.1 2.1
4 0.27 0.13 12 8.2

that yaw angle was an unfavorable effect. As the yaw angle increased,
the thickness of the smoke layer increased.
The flow fields at different yaw angles were shown in Fig. 16. As the
wind yaw angle increased, the longitudinal wind velocity at the tunnel
portal increased. When β = 0◦ , the maximum velocity at the tunnel
portal was 0.75 m/s. When β = 30◦ and β = 60◦ , the maximum velocities
were 0.9 m/s and 1.2 m/s. As the wind yaw angle increased, the lon­
Fig. 14. The temperature rise at different wind velocities. gitudinal velocity component of canyon wind increased and the wind
velocity inside the tunnel increased. Although this effect usually

9
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Fig. 15. The distribution and flow lines of the smoke at different yaw angles.

occurred only to a small length near the tunnel portal, it did lead to an There were fifteen critical conditions obtained from the experiment.
increased risk of smoke diffusing into the tunnel. When the wind yaw angles were 0◦ and 30◦ , the relationship between
Here again, the temperature variation at the tunnel vault was used to canyon wind velocity and critical smoke diffusion length was shown in
respond to smoke diffusion, as shown in Fig. 17 and its error statistics Fig. 18. Took the yaw angle of 0◦ as an example. When the canyon wind
was shown in Table 5. At the yaw angle of 0◦ , ΔTmax inside the tunnel velocity was 0 m/s (no canyon wind influence), the smoke diffusion
was 2.37 ◦ C. When the yaw angle was 30◦ and 60◦ , ΔTmax inside the length was 15 m. When the smoke diffusion length (L) was 12.5 m, 10 m,
tunnel was 8.62 ◦ C and 14.42 ◦ C respectively, which was increased by 7.5 m and 5 m, the critical wind velocity (v) increased 0.9 m/s, 2 m/s,
5.8 ◦ C and 12.05 ◦ C. As canyon wind yaw angle increased, the tem­ 3.1 m/s and 4.6 m/s, respectively. Apparently, the critical wind velocity
perature inside the tunnel increased. (v) decreased as the length (L) increased. The fitting results indicated
In fact, smoke diffusion in the open section was the result of the that L and v might be consistent with an exponential decline. In addition,
competition between thermal buoyancy and wind. As the wind yaw comparing the fitted curves at different wind yaw angles, the two were
angle increased, the component of wind velocity in the open section very similar in shape. It suggested that the fitted curves at different wind
increased in the longitudinal direction and decreased in the transverse yaw angles conformed to the same functional form. When L was 15 m,
direction. Both of the above variations led to an increase in the longi­ the relationship between the wind yaw angles (β) and the critical wind
tudinal influence length of the smoke in the open section. In conclusion, velocity (v) was shown in Fig. 19. When the wind yaw angles (β) were
as the yaw angle increased, the thickness of the smoke layer and the 15◦ , 30◦ ,45◦ and 60◦ , the critical wind velocity (v) increased 0.9 m/s, 2
temperature inside the tunnel increased. When the yaw angle increased m/s, 4.1 m/s and 8 m/s, respectively. The critical wind velocity (v)
by 30◦ , the thickness of the smoke layer increased by 4.73 m m~9.33 m increased as the yaw angles (β) increased. The fitting showed that β and
and the temperature increased by 5.8 ◦ C–12.05 ◦ C. v were consistent with an exponential growth. Because the fire flame and
the wind velocity were not absolutely stable but had certain fluctua­
tions, which resulted in some error in the experiments. That was why
3.5. Analysis of critical conditions for smoke entering the tunnel some data points deviated from the fitted curve. And the experimental
results showed that the error of the critical wind velocity was less than
Typically, when a fire train stops at an open section in the tunnel 0.5 m/s.
group, people need to be evacuated into the tunnel. So, whether the
tunnel is affected by smoke is significant. There was a critical situation, 4. Prediction method of smoke diffusion length affected by
where the smoke diffused at the tunnel portal but didn’t enter the tun­ canyon wind
nel, as shown in Fig. 15(a). In the critical situation the spacing between
the fire source and the tunnel portal was equal to the smoke diffusion In general, modeling experiments was a very reliable research
length (L). Similarly, Hu et al. [57–59] also studied this critical situation method. Due to the high cost of scale experiment, further numerical
in the city street canyon. The effects of the smoke diffusion length (L) on simulations were carried out below to improve the quantitative study of
the critical wind velocity (v) under different wind yaw angle (β) were critical conditions. In view of the fact that FDS have been widely used
analyzed below.

10
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Fig. 16. Flow field at different yaw angles.

and verified in many applications [60], FDS was adopted in this study. boundaries. The tunnel model built in FDS was shown in Fig. 20. During
the FDS simulations, it was assumed that the ambient temperature inside
and outside was the same as the measurement in the tests. The pressure
4.1. Numerical model setup was 94500 Pa. The heat release rate of the fire source was set to 20 MW.
A non-dimensional expression D*/δx was used to measure how well
The tunnel group consisted of three parts: the left tunnel, the open the fire induced flow field could be resolved, which was suggested in
section and the right tunnel. It was built to actual size. The tunnel group FDS user’s guide. Where D* is a characteristic fire diameter (m) and δx is
was uphill from left to right, with a longitudinal slope of 30‰. In order the nominal size of a mesh cell (m). The quantity D*/δx can be thought
to simulate the longitudinal natural wind in the tunnel, the left and right as the number of computational cells span-ning the characteristic
portals of the tunnel group were set as velocity boundaries. The wind diameter of the fire, and the value of D*/δx is ranged from 4 to 16 [61].
velocity was 2.5 m/s and the direction was uniformly uphill. As well, D* is ex-pressed as follows:
one side boundary of the canyon was set as the velocity boundary, which
serveed as the canyon wind inlet. The other side, the upper and lower
boundaries of the canyon were set as “OPEN”, indicating free entry/exit

11
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

( )2/5
Q
D∗ = (11)
ρ0 cp T0 g1/2

where: Q is the heat release rate of fire source, W. ρ0 is the ambient air
density, kg/m3. cp is the specific heat capacity of ambient air, J/(kg⋅K).
T0 is the ambient air temperature, K. g is the gravitational acceleration,
m/s2. It was calculated that the suggested mesh size was about 0.3 m. In
this study, X, Y and Z represented the lateral, longitudinal and vertical
direction respectively. Three different grid sizes (Table 6) were selected
to perform the mesh sensitivity analysis. In the mesh sensitivity analysis,
the length of the open section was 10 m, the canyon wind velocity and
yaw angle were 2 m/s and 30◦ . Fig. 21 showed the distribution of
temperature along longitudinal direction at the height of tunnel vault (9
m) with different grid sizes. Considering the computational expense and
Fig. 17. The temperature rise at different yaw angles. the accuracy comprehensively, the MESH2 was used to conduct the
simulation.

Table 5
4.2. Simulation cases
The temperature errors at different yaw angles.
Yaw angle (◦ ) Absolute error (◦ C) Relative error (%) In order to study how well the FDS predicted the fire smoke flow in
Maximum Mean Maximum Mean the open section, FDS simulation cases with the same boundary condi­
0 0.11 0.08 15 7.9
tion as the critical conditions in experiments were calculated first, as
30 0.28 0.14 5.1 2.3 shown in Table 7. After the meshes and boundary conditions in FDS
60 0.45 0.21 4.1 1.7 simulations were validated by experimental results, more cases were
done in FDS simulations that considering more open section lengths (15
m, 25 m), yaw angles (15◦ ,45◦ ), wind velocities, as shown in Table 8.

4.3. Validation of numerical simulation

Numerical simulations of the same cases were carried out and their
results were compared with the experimental ones. Comparisons in
terms of flow field, temperature and smoke diffusion critical conditions
were examined.
Comparison of the flow field at different canyon wind speeds and
wind yaw angles was shown in Figs. 22 and 23. The velocities inside the
tunnel yielded by simulation were almost equal to 2.5 m/s. That meaned
canyon winds wouldn’t significantly alter the flow field inside the tun­
nel. But as the wind yaw angle increased, the local wind velocity at the
tunnel portal increased. The same conclusion had been obtained in
experimrnts. In addition, it can be seen that the direction of the velocity
vectors obtained from the experiments and simulations was basically the
same. This proved that the flow field results from numerical simulations
Fig. 18. Relationship between critical smoke diffusion length and canyon were reliable.
wind velocity. The temperature variations at the tunnel vault under defferent yaw
angles, open section length, canyon wind velocities were shown in
Fig. 24(a)~Fig. 24(c). Typically, the simulated temperatures were
greater than the experiments. It was due to difference in the thermal
conductivity of the materials. The experimental tunnel walls were made
of metal and the numerical simulations’ wall material was set as con­
crete. In contrast, there was a larger difference in temperature within a
distance of less than 20 m from the portal. This was due to the strong
convection at the tunnel portal, which amplified the effect of differences
in thermal conductivity. The simulations and experiments had the same
laws and similar results. Li (2017) [6] also studied the effect of tunnel
spacing on fires in the open section of the tunnel group and compared
the results of modeling experiments and numerical simulations, as
shown in Fig. 24(d). It showed a strong agreement between simulations
and experiments.
The error analysis of the experimental and simulation results was
shown in Fig. 25. In general, the experimental results were slightly
larger than the simulation results. For the cases, which wind yaw angles
were 0◦ and 3◦ , the above two results were very close to each other, with
Fig. 19. Relationship between critical canyon wind velocity and yaw angle. a standard deviation of less than 0.5. For the wind yaw angle of 60◦ , the
standard deviation of the above two was about 1.0. This was because,
the critical canyon wind velocity was higher at a wind yaw angle of 60◦ .

12
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Fig. 20. Numerical model.

Table 6 Table 8
Grid system sensitivity study. Expansion cases of numerical simulation.
Domain Mesh size(m) Sum NO. D (m) β (◦ ) vc (m/s)

Open section Two tunnels 1~3 10 15 3,3.5,4


4~6 45 0.5,1,1.5
MESH1 0.4(x) × 0.4(y) × 0.2(z) 0.4(x) × 0.4(y) × 0.2(z) 5,436,000
7~9 60 0,0.5,1
MESH2 0.2(x) × 0.2(y) × 0.1(z) 0.4(x) × 0.4(y) × 0.2(z) 5,472,000
10~12 15 0 5.5,6,6.5
MESH3 0.2(x) × 0.2(y) × 0.1(z) 0.2(x) × 0.2(y) × 0.1(z) 10,872,000
13~15 15 4,4.5,5
16~18 30 3,3.5,4
19~21 45 1.5,2,2.5
22~24 60 0.5,1,1.5
25~27 20 15 5,5.5,6
28~30 45 2.5,3,3.5
31~33 60 1.5,2,2.5
34~36 25 0 8.5,9,9.5
37~39 15 6,6.5,7
40~42 30 5,5.5,6
43~45 45 3.5,4,4.5
46~48 60 3,3.5,4
49~51 30 15 10.5,11,11.5
52~54 45 6.5,7,7.5

4.4. Establishment of prediction method

Whether or not the fire smoke in open section diffuses into the tunnel
depends on the competition between the thermal buoyancy and the
wind. This competition can be characterized by the Froude number (Fr)
[57].
v
Fr = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ (12)
Fig. 21. Distribution of temperature along longitudinal direction. (ΔTz /T0 )gz

where: v is the wind velocity, m/s. T0 is the ambient temperature, K. ΔTz


Table 7 is the difference between the temperature at z and the ambient tem­
Validation cases of numerical simulation.
perature, K. g is the gravitational acceleration, m/s2. z is the charac­
NO. D (m) β (◦ ) vc (m/s) teristic height, m. In critical conditions, the Froude number can be
1~3 10 0 4,4.5,5 considered as a constant. It indicates:
4~6 30 1.5,2,3 √̅̅̅̅̅̅̅̅̅̅̅̅̅
7~9 20 0 7,7.5,8 v∞ ΔTH H (13)
10~15 30 0,2,4,4.5,5,6
16 60 2 where: v is the wind velocity in the canyon at the critical condition of
17~19 30 0 14,14.5,15 smoke entry into the tunnel, m/s. H is the height of the tunnel vault, m.
20~23 30 2,8,8.5,9
24~26 60 6,6.5,7
ΔTH is the difference between the temperature at H and the ambient
temperature, K. The height of the tunnel vault is determined by the type
of tunnel section, and ΔTH is determined by the heat release rate of the
The canyon wind in the experiment was driven by axis fans. The fans fire(Q) and the height H. Therefore, the critical wind velocity in the
only drived the air flow in the experiment section and the outside air canyon is related to the type of the tunnel section and the heat release
remained static. Outside air had a large sticky effect on the canyon wind. rate of the fire. Hu et al. [57–59] had studied the critical recirculation
At the same time, when the tunnel spacing was 30 m, the restraining wind velocity of fire smoke in urban street canyons. After careful anal­
effect of canyon winds on the canyon walls was weakened. The above ysis of conservation of the energy, mass and momentum of the fire
two factors made the canyon wind slowdown that cannot be ignored. smoke plume, the relationship between the critical wind velocity and the
And the wind flow field in the actual canyon was extremely large, with HRR of the fire source and street height was established, which also
wind velocities barely decaying in the open section. The velocity applied to this study:
boundary in FDS were more consistent with the actual. In summary,
using FDS to study fires in the open section of the tunnel group under v∞Q1/3 H − 1/3
(14)
canyon wind was very reliable. Such a relationship has been widely recognized [57,58,62]. So the
wind velocity can be dimensionlessized as:

13
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Fig. 23. Comparison of flow fields at defferent yaw angles.


Fig. 22. Comparison of flow fields at defferent canyon wind velocities.

( )1/3 L
v ρcp T0 (16)
v∗ = (15) L∗ =
H
Q1/3 H − 1/3 g
Fig. 26 presented the variation of dimensionless critical spacing with
where: ρ is the ambient air density, kg/m3. cp is the specific heat capacity dimensionless critical wind velocity for different wind yaw angles.
of ambient air, J/(kg⋅K). The variations of L* with v* at different yaw angles fitted by expo­
In the study of pollutant runoff in tunnel groups, Shaw et al. (2015) nential function, logarithmic function and power function were shown
[63] combined the tunnel equivalent diameters to dimensionlessize the in Fig. 26. And their function formula and correlations were shown in
tunnel spacing. As for the study of fires in tunnel groups, Zhao et al. Table 9. Among them, the fitness of exponential function and logarith­
(2021) [7] dimensionlessly normalized the tunnel spacing by the tunnel mic function was higher. And the pattern of the fitted parameters’
section height. Hu et al. (2012) [59] also achieved dimensionlessness by variation in exponential functions was more regular, in particular, there
the aspect ratio of the street in a fire study in a street canyon. Therefore, was only one parameter(A) varying with the yaw angle. Besides, there
the dimensionless spacing between fire source and tunnel portal in this was a strong correlation between smoke diffusion and temperature and
paper was as follow: many scholars usually used the exponential function to describe the
decay of temperature in the tunnel [50,64–66]. Thus, the equation for v*

14
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Fig. 24. Comparison of simulated and experimental point temperatures.

critical case, the fire smoke influence length in the open section was
equal to the spacing between the fire source and the tunnel portal. There
was a global correlation function for the critical smoke influence length
as following:
( ∗ )
v + 2.5
L∗ = − 1.63 ln (18)
2e1.5β + 5

4.5. Error analysis

In order to validate the reliability of the prediction method estab­


lished through numerical simulation, the experimental results were used
for error analysis. The comparison of the predicted and experimental
results was shown in Table 10. The maximum absolute error of the
predicted results and the experimenral results was 0.94 m/s and the
maximum relative error was 6.69%. There was a good agreement be­
tween the prediction and the experimental results. Therefore, the pre­
Fig. 25. Comparison of simulated and experimental critical conditions. diction method was reliable.

and L* at different yaw angles was established as: 5. Conclusions



v = Ae − 0.61L∗
− 2.5 (17) In this paper, the effects of canyon winds on fire smoke diffusion
were studied by means of literature survey, modeling experiment, nu­
where:A is a parameter associated with the wind yaw angle. As the wind
merical simulation and theoretical analysis. A prediction model of
yaw angle increased, only the parameter A increased. Therefore, we
smoke diffusion length was established. The conclusions were as
considered the wind yaw angle as an independent variable, which only
follows:
affected the parameter A. The parameter A at different wind yaw angles
was shown in Fig. 27, where angles were converted to radians. In the

15
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Fig. 26. Dimensionless critical smoke influence length against canyon wind velocity.

(1) Wind velocity and yaw angle are the major characteristics of
Table 9 canyon wind that affected smoke diffusion. The suggested values
Comparison of fitted functional forms. of wind velocities and yaw angles are 0~15 m/s and 0–60◦ ,
Function form β(◦ ) Formula R2 Mean R2 respectively.
− 0.61x (2) As canyon wind velocity(v) increases, the fire smoke diffusion
Exponential function 0 y = 7.14e -2.5 0.95 0.948
15 y = 8.33e− 0.61x-2.5 0.94 length in the open section (L) decreases. When L is less than the
30 y = 9.50e − 0.61x
-2.5 0.96 distance of the fire source from the tunnel portal (D), the smoke
45 y = 10.74e− 0.61x-2.5 0.93 wouldn’t endanger the passengers’ evacuation in the tunnel.
60 y = 14.88e− 0.61x-2.5 0.96
When L is greater than D, the dangerous smoke would enter the
Logarithmic function 0 y = 3.2–3.79ln (x+0.63) 0.95 0.950
15 y = 22.64–13.23ln (x+3.74) 0.97
tunnel. And greater v helps to reduce smoke temperature as well
30 y = 4.39–4.45ln (x+0.49) 0.96 as the thickness of the smoke layer at the tunnel portal.
45 y = 2.58–2.07ln (x-0.25) 0.93 (3) As the yaw angle (β) increases, the length of fire smoke diffusion
60 y = 4.2–2.66ln (x-0.32) 0.94 (L) increases in the open section. The yaw angle works opposite of
Power function 0 y = 7.17–5.82x0.39 0.93 0.930
canyon wind velocity. As the smoke enters the tunnel, greater β
15 y = 5.58–3.51x0.79 0.92
30 y = 11.02–8.41x0.35 0.94 makes smoke temperature and smoke layer thickness higher and
45 y = 13.58–10.27x0.29 0.94 thicker, which is unfavorable.
60 y = 18.67–13.14x0.34 0.92 (4) A prediction method of smoke diffusion length affected by
canyon winds is established, which error is less than 10%. It could
determine whether the fire smoke would endanger passengers’
evacuation in case of natural diffusion, and provide a basis for
setting up smoke control ventilation.

In this study, we ignored the variation of natural winds in the tun­


nels. Didn’t take defferent side slope angles and longitudinal slopes into
account. In addition, canyon winds were simplified to a steady wind
flow and some of its properties were neglected, such as pulsatility. In the
future, we will further study on the factors mentioned above in the
detail. And the prediction method will also be improved to be more
applicable.

CRediT authorship contribution statement

Xin Chen: Data curation, Formal analysis, Writing – original draft. Li


Yu: Methodology, Writing – review & editing. Jianxun Huo: Resources,
Validation, Visualization. Xue Wang: Software, Supervision. Songshen
Fig. 27. The parameter A against canyon wind yaw angle. Wang: Software. Xiaohan Guo: Methodology, Project administration.
Mingnian Wang: Project administration, Writing – review & editing.

Table 10
Error analysis of prediction method.
NO. Yaw angle (◦ ) Canyon wind velocity (m/s) Predicted result(m) Experimental result(m) Absolute error(m) Relative error (%)

1 0 0 15.94 15 0.94 6.29


2 0 2 10.02 10 0.02 0.21
3 0 4.8 4.67 5 0.33 6.69
4 30 2 14.56 15 0.44 2.91
5 30 4.6 9.74 10 0.26 2.60
6 30 7.9 4.84 5 0.16 3.25

16
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

Declaration of competing interest [26] S. Miao, H. Xiong, D. Li, Y. Gu, A two-phase wind speed simulation model
considering diurnal and seasonal patterns and its application to adequacy
assessment, Journal of Renewable and Sustainable Energy 13 (5) (2021).
The authors declare that they have no known competing financial [27] F. Jiang, M. Zhang, Y. Li, T. Yan, J. Zhang, Field measurement analysis of wind
interests or personal relationships that could have appeared to influence parameters and nonstationary characteristics in mountainous terrain: Focusing on
the work reported in this paper. cooling windstorms, Journal of Wind Engineering and Industrial Aerodynamics
230 (2022) 105175.
[28] M. Zhang, F. Jiang, Y. Li, H. Chen, J. Qin, L. Wu, Multi-point field measurement
Data availability study of wind characteristics in mountain terrain: focusing on periodic thermally-
developed winds, Journal of Wind Engineering and Industrial Aerodynamics 228
(2022) 105102.
No data was used for the research described in the article. [29] A. Fenerci, O. Øiseth, A. Rønnquist, Long-term monitoring of wind field
characteristics and dynamic response of a long-span suspension bridge in complex
References terrain, Engineering Structures 147 (2017) 269–284.
[30] Y. Li, F. Jiang, M. Zhang, Y. Dai, J. Qin, J. Zhang, Observations of periodic
thermally-developed winds beside a bridge region in mountain terrain based on
[1] N.R.A.o.t.P.s.R.o. China, Code for Design on Rescue Engineering for Disaster
field measurement, Journal of Wind Engineering and Industrial Aerodynamics 225
Prevention and Evacuation of Railway Tunnel, China Railway Press, Beijing, 2017.
(2022) 104996.
[2] I.U.o. Railways, Safety in Railway Tunnels, UIC, Paris, 2002.
[31] D.F. Castellon, A. Fenerci, O. Øiseth, Environmental contours for wind-resistant
[3] C. Regulation, Concerning the Technical Specification for Interoperability Relating
bridge design in complex terrain, Journal of Wind Engineering and Industrial
to ’safety in Railway Tunnels’ of the Rail System of the European Union, Official
Aerodynamics 224 (2022) 104943.
Journal of the European Union, Brussels, 2014.
[32] W. Yang, Y. Liu, E. Deng, Y. Wang, X. He, M. Lei, Characteristics of wind field at
[4] S.I. Association, Design of Tunnels Railway Tunnels, Swiss Society of Engineers and
tunnel-bridge area in steep valley: field measurement and LES study, Measurement
Architects, Zurich, 2004.
202 (2022) 111806.
[5] F.K. Chow, S.F. De Wekker, B.J. Snyder, Mountain Weather Research and
[33] D.-H. Yang, T.-H. Yi, H.-N. Li, Y.-F. Zhang, Monitoring-based analysis of the static
Forecasting: Recent Progress and Current Challenges, Springer2013..
and dynamic characteristic of wind actions for long-span cable-stayed bridge,
[6] L. Qi, Research on Reliability of Personnel Safe Evacuation under High-Speed
Journal of civil structural health monitoring 8 (2018) 5–15.
Railway Super-long Tunnel in Fire, Southwest Jiaotong University, Chengdu, 2018.
[34] C. Ma, C. Pei, H.-l. Liao, M. Liu, M. Li, Field measurement and wind tunnel study of
[7] Z. Zunxin, T. Fei, Z. Yuantao, Study on the characteristics of smoke crossflow of the
aerodynamic characteristics of twin-box girder, Journal of Wind Engineering and
tunnel group in tunnel fire, Journal of Engineering Thermophysics 11 (2021) 42.
Industrial Aerodynamics 202 (2020) 104209.
[8] W. Haoran, Research on the Smoke Flow Mechanism of the Fire Train Entering the
[35] C. Liu, Y. Gong, Z. Jiang, K. Guo, Buffeting analysis of a suspension bridge under
Rescue Station at the Tunnel Portals, Southwest Jiaotong University, 2021.
construction based on adjacent wind field data, Engineering Structures 251 (2022)
[9] M.o.T.o.t.P.s.R.o. China, Wind-resistant Design Specification for Highway Bridges,
113490.
China Communications Press, 2018.
[36] N.A.O. Sciences, N. Council, Fire Research Abstracts and Reviews ume 11 (1969).
[10] E.C.F. Standardization, Eurocode 1:Actions on Structures, Part 1-4, General
[37] N. Ikegaya, H. Kikumoto, K. Sasaki, S. Yamada, M. Matsui, Applications of wide-
actions-Wind actions, Brussels, 2006.
ranging PIV measurements for various turbulent statistics in artificial atmospheric
[11] F. Jiang, M. Zhang, Y. Li, J. Zhang, J. Qin, L. Wu, Field measurement study of wind
turbulent flow in a wind tunnel, Building and Environment 225 (2022) 109590.
characteristics in mountain terrain: Focusing on sudden intense winds, Journal of
[38] Y. Zhao, Y. Xue, S. Mei, Y. Chao, J. Carmeliet, Enhancement of heat removal from
Wind Engineering and Industrial Aerodynamics 218 (2021) 104781.
street canyons due to buoyant approaching flow: water tunnel PIV-LIF
[12] M. Laib, J. Golay, L. Telesca, M. Kanevski, Multifractal analysis of the time series of
measurements, Building and Environment 226 (2022) 109757.
daily means of wind speed in complex regions, Chaos, Solitons & Fractals 109
[39] P. Farias, L. Azevedo, I. De Paula, Characterization of interfacial waves in stratified
(2018) 118–127.
turbulent gas-liquid pipe flow using Particle Image Velocimetry and controlled
[13] F. Jiang, J. Zhang, M. Zhang, Y. Li, J. Qin, Field measurement study on
disturbances, International Journal of Multiphase Flow 161 (2023) 104381.
classification for mixed intense wind climate in mountainous terrain, Measurement
[40] N. Ikegaya, S. Hasegawa, A. Hagishima, Time-resolved particle image velocimetry
217 (2023) 113064.
for cross-ventilation flow of generic block sheltered by urban-like block arrays,
[14] H. Jentsch, J. Weidinger, Spatio-temporal analysis of valley wind systems in the
Building and Environment 147 (2019) 132–145.
complex mountain topography of the Rolwaling Himal, Nepal, Atmosphere 13 (7)
[41] P. Górski, S. Pospíšil, M. Tatara, A. Trush, PIV analysis of near-wake flow patterns
(2022) 1138.
of an ice-accreted bridge cable in low and moderately turbulent wind, Journal of
[15] J. Zhang, M. Zhang, Y. Li, F. Jiang, L. Wu, D. Guo, Comparison of wind
Wind Engineering and Industrial Aerodynamics 191 (2019) 297–311.
characteristics in different directions of deep-cut gorges based on field
[42] F. Scarano, Overview of PIV in supersonic flows, in: Particle Image Velocimetry:
measurements, Journal of Wind Engineering and Industrial Aerodynamics 212
New Developments and Recent Applications, 2007, pp. 445–463.
(2021) 104595.
[43] Y. Ozawa, T. Ibuki, T. Nonomura, K. Suzuki, A. Komuro, A. Ando, K. Asai, Single-
[16] W. Zhao, R. Ouyang, Q. Ran, T. Chen, Z. Xu, M. Zou, C. Fan, An experimental study
pixel resolution velocity/convection velocity field of a supersonic jet measured by
on smoke back-layering and critical velocity in tunnel fires with canyon cross wind,
particle/schlieren image velocimetry, Experiments in Fluids 61 (2020) 1–18.
Journal of Wind Engineering and Industrial Aerodynamics 209 (2021) 104477.
[44] Y. Zhuang, Z. Liu, W. Xu, Effects of gradient porous metal foam on the melting
[17] D. Luan, L. Yi, L. Yang, T. Chen, H. Tao, Z. Xu, C. Fan, Experimental investigation
performance and energy storage of composite phase change materials subjected to
of smoke temperature and movement characteristics in tunnel fires with canyon
an internal heater: a numerical study and PIV experimental validation,
cross wind, Journal of Wind Engineering and Industrial Aerodynamics 210 (2021)
International Journal of Heat and Mass Transfer 183 (2022) 122081.
104531.
[45] P. Wei, Effect of Longitudinal Ventilation on Burning and Smoke Movement in
[18] Y. Shen, A. Jiao, T. Chen, Y. Li, Y. Gao, Z. Xu, B. Jiang, C. Fan, Experimental study
Road Tunnel Fires, University of Science and Technology of China, 2008.
on smoke movement characteristics in tunnel fires with different canyon cross
[46] D. Oettl, P.J. Sturm, M. Bacher, G. Pretterhofer, R.A. Almbauer, A simple model for
wind yaw angles, Tunnelling and Underground Space Technology 117 (2021)
the dispersion of pollutants from a road tunnel portal, Atmospheric Environment
104129.
36 (18) (2002) 2943–2953.
[19] A. Jiao, Y. Shen, Z. Wang, T. Chen, H. Tao, Z. Xu, C. Fan, Experimental study on the
[47] Y. Zhou, H. Bi, H. Wang, B. Lei, Critical velocity in the transverse passages of a
effect of canyon cross wind yaw angle on airflow and flame characteristics in a
railway tunnel rescue station with semi-transverse ventilation, Tunnelling and
tunnel, Journal of Wind Engineering and Industrial Aerodynamics 213 (2021)
Underground Space Technology 92 (2019) 103064.
104616.
[48] M.o.T.o.t.P.s.R.o. China, Guidelines for Design of Ventilation of Highway Tunnels,
[20] C. Yu, Y. Li, M. Zhang, Y. Zhang, G. Zhai, Wind characteristics along a bridge
China Communications Press, Beijing, 2014.
catwalk in a deep-cutting gorge from field measurements, Journal of Wind
[49] L. Hu, S. Liu, W. Peng, R. Huo, Experimental study on burning rates of square/
Engineering and Industrial Aerodynamics 186 (2019) 94–104.
rectangular gasoline and methanol pool fires under longitudinal air flow in a wind
[21] T.M. Lystad, A. Fenerci, O. Øiseth, Evaluation of mast measurements and wind
tunnel, Journal of hazardous materials 169 (1–3) (2009) 972–979.
tunnel terrain models to describe spatially variable wind field characteristics for
[50] Y.-c. Zhao, G.-q. Zhu, Y.-j. Gao, Experimental study on smoke temperature
long-span bridge design, Journal of Wind Engineering and Industrial Aerodynamics
distribution under different power conditions in utility tunnel, Case studies in
179 (2018) 558–573.
thermal engineering 12 (2018) 69–76.
[22] H. Xinmin, G. Wenhua, X. Anping, Numerical simulation of distribution
[51] Y. Wang, J. Jiang, D. Zhu, Full-scale experiment research and theoretical study for
characteristic of wind fields and terrain’s influence in mountain canyon, Journal of
fires in tunnels with roof openings, Fire Safety Journal 44 (3) (2009) 339–348.
Chang’an University(Natural Science Edition) 37 (5) (2017) 56–64.
[52] Y. Kunikane, Heat release rate induced by gasoline pool fire in a large-cross-section
[23] P.L. Jackson, G. Mayr, S. Vosper, Dynamically-driven Winds, Mountain weather
tunnel, Tunnel Management International 6 (4) (2003) 22–29.
research and forecasting: Recent progress and current challenges, 2013,
[53] V. Apte, A. Green, J. Kent, Pool fire plume flow in a large-scale wind tunnel, Fire
pp. 121–218.
Safety Science 3 (1991) 425–434.
[24] E. Simiu, R.H. Scanlan, Wind Effects on Structures: Fundamentals and Applications
[54] N. Kang, Y. Qin, X. Han, B. Cong, Experimental study on heat release rate
to Design, John Wiley New York1996..
measurement in tunnel fires, Fire and materials 43 (4) (2019) 381–392.
[25] H.R. Byers, H. Landsberg, H. Wexler, B. Haurwitz, A. Spilhaus, H. Willett, H.
[55] L. Li, Research on the Recognition Algorithm for Concentration and Diameter
Houghton, Compendium of Meteorology, Springer1951..
Distribution of Indoor Suspended Particulate, Wuhan University of Technology,
2008.

17
X. Chen et al. International Journal of Thermal Sciences 200 (2024) 108957

[56] C.G. Fan, J. Yang, Experimental study on thermal smoke backlayering length with [61] K.B. McGrattan, G.P. Forney, Fire Dynamics Simulator: User’s Manual, US
an impinging flame under the tunnel ceiling, Experimental Thermal and Fluid Department of Commerce, Technology Administration, National Institute of, 2000.
Science 82 (2017). [62] Q. Wang, T. Zhou, Q. Liu, P. He, C. Tao, Q. Shi, Numerical study of critical re-
[57] L. Hu, Y. Xu, W. Zhu, L. Wu, F. Tang, K. Lu, Large eddy simulation of pollutant gas entrainment velocity of fire smoke within the street canyons with different building
dispersion with buoyancy ejected from building into an urban street canyon, height ratios, Environmental Science and Pollution Research 26 (2019)
Journal of hazardous materials 192 (3) (2010) 940–948. 23319–23327.
[58] L. Hu, R. Huo, D. Yang, Large eddy simulation of fire-induced buoyancy driven [63] X. Yi-min, P. Xiao-hong, Z. Jin-peng, Model test system design of channeling effects
plume dispersion in an urban street canyon under perpendicular wind flow, of air pollutants in continuous road tunnels, China Journal of Highway and
Journal of hazardous materials 166 (1) (2009) 394–406. Transport 28 (11) (2015) 90.
[59] L. Hu, X. Zhang, W. Zhu, Z. Ning, F. Tang, A global relation of fire smoke re- [64] M.A. Delichatsios, The flow of fire gases under a beamed ceiling, Combustion and
circulation behaviour in urban street canyons, Journal of Civil Engineering and Flame 43 (1981) 1–10.
Management 21 (4) (2012) 459–469. [65] J. Bailey, G.P. Forney, P. Tatem, W.W. Jones, Development and validation of
[60] J. Trelles, J.R. Mawhinney, CFD investigation of large scale pallet stack fires in corridor flow submodel for CFAST, Journal of Fire Protection Engineering 12 (3)
tunnels protected by water mist systems, Journal of fire protection engineering 20 (2002) 139–161.
(3) (2010) 149. [66] M.B. Kim, Y.S. Han, M.O. Yoon, Laser-assisted visualization and measurement of
corridor smoke spread, Fire safety journal 31 (3) (1998) 239–251.

18

You might also like