Journal of Natural Gas Science and Engineering 84 (2020) 103636

Contents lists available at ScienceDirect

Journal of Natural Gas Science and Engineering

journal homepage: http://www.elsevier.com/locate/jngse

Location of natural gas leakage sources on offshore platform by a

multi-robot system using particle swarm optimization algorithm
Tao Ma a, Shuhai Liu a, b, *, Huaping Xiao a, b
a
College of Mechanical and Transportation Engineering, China University of Petroleum-Beijing, Beijing, 102249, China
b
Laboratory of Petroleum and Petrochemical Robot, China University of Petroleum-Beijing, Beijing, 102249, China

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

Keywords: The offshore platform serves as the infrastructure for offshore mining. In order to guarantee safe operation of the
Offshore platform platform and the development of the platform in the direction of intelligence, a natural gas leakage intelligent
Natural gas leak detection system with fast response is necessary to prevent potential accident when leakage occurs. In this work,
Particle swarm optimization algorithm
a method to locate leaked natural gas is proposed based on the particle swarm optimization algorithm for the
Gas source search
multi-robots to achieve. First, the advection diffusion equation is calculated to simulate the leakage gas trans­
mission using finite difference method in different scene layouts. At the same time，a multi-robot collaborative
detection strategy for gas detection, gas tracking, and gas source localization is proposed using an improved
particle swarm optimization algorithm. After that, the process of multi robots search for leak sources was
simulated. In addition, the control algorithm is analyzed, and the results show that the strategy has a high
detection rate. Further analysis shows that the number of robots has increased from 2 to 7, the time it takes for
the multi-robot system to successfully locate the leak source decreased; when the number of robots is constant,
the time it takes for the multi-robot to successfully locate two leak sources is longer than the time to locate a
single leak. Finally, search strategies based on particle swarm optimization, ant colony algorithm and cuckoo
search algorithm are used to simulate in a single leak source environment. When the number of robots is changed
from 3 to 7, compared with the other two algorithms, the accuracy of the search strategy search based on particle
swarm is increased by 2%–28%, and the search time is accelerated by 27s–92s.the particle swarm optimization is
superior than two others for control of multi-robot system in leakage source location.

1. Introduction triggering Domino effect (Almeida and Vinnem, 2020; Kim et al., 2016.
In July 1988, due to inadvertent personnel operation, the combustible
With the increasing demand for energy in the development of society gas leak of the compressor module caused an explosion on the Piper
and the depletion of onshore oil and gas resources, the development of Alpha platform, killing 167 workers and causing economic losses of up
oil and gas resources in various countries has gradually shifted to the to 28.7 pounds (Shallcross et al., 2013). In April 2010, due to a hydro­
development of offshore oil and gas resources. As an important infra­ carbon leak, the “Deepwater Horizon” platform exploded, killing 11
structure for offshore oil and gas resource exploitation and production, people and causing a large-scale marine oil spill in the Gulf of Mexico
the stability of the platform is an important condition to ensure normal (Yang et al., 2019a). In 2012, the pipeline rupture due to corrosion, a
exploitation (Bao et al., 2020). The offshore platform has compact natural gas leak occurred at the French company Total Oil’s “Elgin”
structure, small activity space and high personnel density. The materials offshore platform in the North Sea off Scotland.
have the characteristics of high temperature and high pressure, At present, natural gas leak detection technologies are based on
inflammable and explosive, toxic and harmful, and continuous exis­ acoustic, temperature, artificial neural network, etc. The common
tence. Not only that, the marine environment where the platform is detection technology for leaking gas on the offshore platforms is mainly
located is complex and changeable, and fire and explosion accidents are sensor network monitoring technology (Richart et al., 2016). When the
prone to occur due to natural gas leakage and other reasons (Liu et al., diffused gas is detected, the sensor alarm. This detection method is
2018; Olsen et al., 2015), causing catastrophic consequences and currently applied to offshore platforms and is constantly being

* Corresponding author. College of Mechanical and Transportation Engineering, China University of Petroleum-Beijing, Beijing, 102249, China.
E-mail address: [email protected] (S. Liu).

https://doi.org/10.1016/j.jngse.2020.103636
Received 15 April 2020; Received in revised form 20 September 2020; Accepted 21 September 2020
Available online 28 September 2020
1875-5100/© 2020 Elsevier B.V. All rights reserved.
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

optimized (Seo et al., 2013), but there are certain disadvantages that probabilistic method to identify partial differential equations under
make the monitoring results not very accurate: a. Due to the complexity sparsity, and used factor graphs and message passing algorithms for
of the marine environment, the detection sensor may generate false multi-robots The process of gas source location and gas source quantity
alarms in high temperature and high humidity environments; b. The identification is simulated, and the effectiveness of the algorithm is
detection sensor has a dead zone, and the gas diffusion process is verified through experiments.
changeable. When the number of sensors is insufficient, gas may not be The smell source location method based on smell or vision locates the
detected; c. The fixed position of the sensor makes it difficult to accu­ smell source by using gas diffusion information obtained by smell or
rately locate the source of the diffused gas leak. Based on the analysis of vision sensors. At present, there are relatively few reports on air source
recorded data from 1992 to 2016, in the offshore oil industry, 42% of gas localization based on this method. Loutfi (Loutfi et al., 2004) uses the
leak accidents (Leak rate is lower than 1 kg/s) were not detected by robot to locate the odor source, uses visual sensors to identify the sus­
on-site gas detectors (Zhou et al., 2019). All in all, the use of sensor pected odor source, and then the robot judges each suspected odor
networks for gas leak detection on offshore platforms deserves to be source. Li Wei (Li, 2007) proposed an iterative fuzzy segmentation al­
reviewed. gorithm to extract the color information of smoke plumes and odor
Based on the above reasons, this paper proposes a scheme for multi- sources by analyzing the underwater lighting conditions in the offshore
robots to collaboratively search for natural gas leakage on the offshore environment based on visual information.
platforms. The physical space of multi-robots consists of a large number The swarm intelligence optimization search algorithm uses swarm
of robots, and the system control method combines a centralized robots to identify and locate the odor source based on a certain algo­
network and a distributed network for control. The centralized network rithm strategy. Due to the update of science and technology and the
effectively eliminates conflicts between multiple robots, and all robots swarm intelligence algorithm has a better search effect and stronger
communicate in real time with the central controller. In decentralized robustness compared with the gas search algorithm, this Algorithms
network, each robot’s decision-making depends on the information of its have received more and more attention from researchers in recent years
own sensors. In this case, each robot analyzes the group information and (Marjovi and Marques 2013, 2014). Common swarm intelligence algo­
its own information and then executes the next action, which improves rithms include PSO (Feng et al., 2018, 2020; Yang et al., 2019a, b; Jain
the robot’s overall problem-solving ability, making it more reliable and et al., 2019), genetic algorithm (Villarreal et al., 2016), ant colony al­
robust (Shukla and Karki, 2016; Marangoz et al., 2019). Compared with gorithm, etc. (Meng et al., 2012; Zou and Luo, 2008; Khalili et al., 2013).
static sensor network monitoring method, multi-robots actively search is Qilin Feng (Feng et al., 2018, 2020; Yang et al., 2019a, b) et al. estab­
more flexible. On the other hand, this paper adopts the multi-robot lished indoor time-varying pollutant source leakage models in
divergent gas searching method and the gas tracking strategy based on two-dimensional and three-dimensional environments, and used PSO to
the particle swarm optimization (PSO) algorithm. Through the optimi­ simulate the multi-robot active olfactory search strategy and verified the
zation and improvement of the PSO algorithm, the control strategy of time the effectiveness of the variable leakage source location method for
the multi-robot system is improved to achieve the detection of leaking source location of pollutants. Based on PSO algorithm and group strat­
natural gas on the platform more efficiently. egy, Upma Jain (Jain et al., 2019) et al. used multiple robots to locate
In recent years, related scholars have done a lot of work and achieved multiple odor sources in an unknown environment. In the process of gas
many results in different methods for the detection and location of positioning, the degree of aggregation of multiple robots is restricted
diffuse gas. Usually we divide the diffuse odor source location method based on a group, and a certain combination behavior is selected by
into four categories according to the different focus, they are a method measuring the precision. At the same time, the proposed framework is
based on bionic behavior, a method based on model analysis, a method evaluated based on the measurement error, and the effectiveness of the
based on smell or vision, and a group intelligent optimization search proposed scheme is verified. B. Lorena Villarreal (Villarreal et al., 2016)
algorithm. This paper will give an overview of the different methods. et al. developed a heuristic robot odor source location algorithm based
The odor source location method based on bionic behavior mainly on genetic algorithm. The quality of results demonstrates that genetic
includes the algorithm based on concentration gradient (Rozas et al., programming is able to recreate chemotaxis behavior.
1991), the lobster heuristic algorithm based on biological behavior The four types of robot air source positioning algorithms are
(Reeder and Ache, 1980; Consi et al., 1994), the silkworm moth heuristic reviewed and analyzed. The multi-robot positioning methods based on
algorithm (Kuwana et al., 1996; Lilienthal et al., 2003), and ZigZag bionic behavior search and visual and olfactory search have received
traversal (Farrel et al., 2003) algorithm, Spiral algorithm (Hayes et al., less attention at present. These two types of methods are more complex
2002) and so on. A single robot searching and locating the source based than swarm intelligence and model-based. The positioning method is
on the algorithm of concentration gradient has been studied earlier. simpler and the use cost is relatively low. However, the multi-robot
Chemotaxis and Biased Random Walk are two kinds of typical localization methods based on bionic behavior search and visual and
gradient-based algorithms. Chris Lytridis (Lytridis et al., 2006) et al. are olfactory search are only applicable to a single robot in most cases, and it
based on chemotaxis, biased random walk and combined chemo-BRW is difficult to form a suitable multi-robot movement strategy, which
strategy for single robot and multi-robot Source search verification, leads to a decrease in the localization efficiency (Hayes et al., 2002;
the results show that the algorithm not only has a good search effect, but Khalili et al., 2013). Intelligent swarm robots and probabilistic
also multi-robots improve the search efficiency and robustness of the model-based robot air source positioning methods are expensive to use,
search strategy. but due to their strong search adaptability, short search time, and high
The method based on model analysis is usually based on wind field or positioning efficiency, they have been paid attention to by a large
plume model analysis, and estimate and predict the location and diffu­ number of research scholars over the years. Compared with other
sion path of odor source. Ali Marjovi (Marjovi and Marques, 2011) and methods, the robot air source positioning method based on the proba­
others proposed an algorithm based on decentralized boundaries, which bility model has more complicated calculations and takes up more CPU,
uses multiple robots to search for odor sources in an unknown structured so it is not easy for some old hardware platforms (Chen et al., 2019).
environment. During the search process, each robot independently This study is based on PSO algorithm to achieve the multi-robot
generates an online local topology map and shares them with each other. searching for the source of natural gas leakage on the offshore plat­
To build a global map. And under different numbers of robots and form. In the second part, the environment of the offshore platform is
different scenarios, tests were carried out in simulation and real envi­ analyzed, and the advection diffusion equation is used to numerically
ronments respectively to verify the effectiveness of the algorithm. simulate the diffusion process of the leaked natural gas on the offshore
Thomas Wiedemann (Wiedemann et al.,2018, 2019) et al. used partial platform. At the same time, based on the traditional PSO algorithm, a
differential equations to model the gas diffusion process, and proposed a strategy for multi-robots to search for the leaked natural gas is proposed.

2
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

Fig. 1. Research framework.

equation (ADE) to simulate natural gas on the platform (Tirabassi et al.,

2009). The ADE belongs to the second-order partial differential equation
and often used in the fields of oil and gas development, fluid mechanics,
environmental engineering, and electronic information science. It used
to describe various natural phenomena such as gas diffusion, heat con­
duction, and pollutant diffusion. The equation consists of two processes:
advection and diffusion. Advection includes the transfer of chemical
substances in a gas. Diffusion indicates that the gas migrates from a high
concentration region to a low concentration region. Commonly used
numerical calculation methods for convection-diffusion equations
include finite volume method (FVM), finite difference method (FDM),
finite element method (FEM), etc. (Wei and Xu, 2016), this research uses
the FDM to numerically calculate the equation. In the calculation pro­
cess, the FDM can be divided into first-order format, second-order
format, and high-order format according to the accuracy of the solu­
tion format, and can be divided into the central difference format and
the upwind difference format according to the spatial pattern, and ac­
cording to the time factor can be divided into explicit format, implicit
format, etc. The leakage gas diffusion process is simulated and expressed
by equation (1).
∂f （x, t）
− kΔf (x, t) = u(x, t) (1)
∂t

Fig. 2. Multi-Robots searches on the offshore platform. where f(x, t) describe the gas concentration at point xand time t. k > 0,
The factor k before the Laplace operator Δf(x, t) is the diffusion rate.
2 2
f(x,t) f(x,t)
The third part uses MATLAB R2018b software to simulate the robot u(x, t) represents gas diffusion source. Δf(x, t) = ∂ ∂x +∂ ∂y is the
search process, and analyzes the impact of different scenarios, the divergence of f(x, t) in two dimensional space.
number of leak sources and the number of different robots on the multi- Implicit difference construction for equation (1):
robot detection effect, and compares the efficiency of PSO, colony ( )
k+1
optimization (ACO), and cuckoo search algorithm (CSA). Finally, the fi,j − fi,jk
k+1
fi+1,j − 2fi,jk+1 + fi−k+1 k+1
fi,j+1 k+1
− 2fi,jk+1 + fi,j−
(2)
1,j 1
− k + = u(x, t)
fourth part is the summary and prospect of the research. The research Δt Δx Δy
framework is shown in Fig. 1. Multi robot searches on the offshore
platform is shown in Fig. 2. Δt is the time step, Δx、Δy are the grid steps in the two-dimensional
space X–Y directions, l is the side length, M is the number of grids, Δx =
2. Method Δy = h = Ml ，the truncation error is Ο(Δx2 + Δy2 + Δt2 ). Let r = kΔt h ,
then formula (2) can be written as:
2.1. Leaked natural gas diffusion model ( )
(1 + 4r)fi,jk+1 − r fi+1,j
k+1
+ fi−k+1 k+1 k+1 k
1,j + fi,j+1 + fi,j+1 = fi,j + Δt*u(x, t) (3)
2.1.1. Gas diffusion equation
The research methods of the diffusion law of leaked gas mainly The difference pattern can be represented by the following matrix:
include field test, wind tunnel experiment and computational fluid dy­ Af n+1 = f n + ϕ (4)
namics (CFD) simulation. This article uses the advection diffusion

3
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

Fig. 3. Gas diffusion model.

Fig. 4. Offshore platform scene model.

where ⎡
A1,1 − rI
⎤
[ ]T ⎢ − rI A2,2 − rI ⎥
f n+1 = f1,1
n+1 n+1 n+1
, f1,2 …f1,M− n+1 n+1 n+1 n+1 ⎢ ⎥
1 , f2,1 …f2,M− 1 …fM− 1,1 …fM− 1,M− 1 A=⎢ ⋱ ⎥
⎣ ⎦
[ ]T − rI AM− 2,M− 2 − rI − rIAM− 1,M− 1
(M− 1)×(M− 1)
f n+1 = f1,1
n n
, f1,2 n
…f1,M− n n n
1 , f2,1 …f2,M− 1 …fM−
n
1,1 …fM− 1,M− 1

where
ϕ = Δt*u(x, t)

find the coefficient matrix A

4
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

Table 2
PSO Parameter initialization.
PSO Parameter initialization

Number of robots m 3
ω 0.8
c1 0.5
c2 0.5
r1 Rand（0,1）
r2 Rand（0,1）
vmax 0.1

⎡ ⎤
1 + 4r − r
⎢ − r 1 + 4r − r ⎥
⎢ ⎥
Ai,j = ⎢ ⋱ ⎥
⎣ ⎦
− r 1 + 4r− r − r1 + 4r
(M− 1)×(M− 1)

⎡ ⎤
1
Fig. 5. Particle moving direction.
⎢ 1 ⎥
⎢ ⎥
I=⎢ ⋱ ⎥
⎣ 1 ⎦
1 (M− 1)×(M− 1)

finally found

f n+1
= A− 1 (f n + ϕ) (5)
It can be known from the general difference theory that the differ­
ence scheme is stable and convergent.

2.1.2. Diffusion simulation

This research uses MATLAB R2018b software simulated the natural
gas diffusion process in a two-dimensional space. The natural gas
diffusion environment is Ω = (0, l) × (0, l) square two-dimensional
space, where l = 2m, pixels are 20 × 20. We assume that a grid size of
the leak source location, set a single leak source environment, the
location is (1, 1); two leak source environments, the location is (0.5, 1.5)
and (1.5, 0.5) respectively. The diffusion coefficient of ADE is k = 0.1,
and the Dirichlet boundary condition is used to solve the convection
diffusion equation by the FDM. Assuming that natural gas continues to
diffuse continuously, the diffusion effect is shown in Fig. 3, the leakage
source concentration is 1.

2.1.3. Offshore platform scene

There are many types of offshore platforms. Due to their different
Fig. 6. Robots diffusion movement.
scales and different deck functions, the layout of the equipment on the
platforms varies. The layout of the equipment has a certain effect on gas
diffusion and robot motion. To more realistically simulate the real sit­
Table 1
uation on the platform, four platform scenarios are set. In Fig. 4, the
Search strategy pseudo code based on PSO algorithm.
yellow area is indicated as barrier-free, and the black area is the device
Initialization parameters; layout.
Robot position initialization
do{
If (No gas found)
{ 2.2. The basic principle of PSO
The robots scattered and set out to find gas
} The PSO emerged in the 1990s. The algorithm simulates bird-
else
{
predation behavior and seeks the optimal solution through collabora­
Start PSO strategy; tion and information sharing among individuals (Das et al., 2016).
Calculate individual extreme values pbest , global extreme values gbest Because the PSO is simple and easy to implement and there are no ad­
for (i = 1; i < m; i++) % m is the number of robots justments for many parameters, it has been widely used in function
{ optimization, neural network training, fuzzy system control, and other
Update position according to formula (6)
genetic algorithm applications. The basic principle is that the bird group
Update speed according to formula (7)
} adjusts its search path through its own experience and the communi­
Update individual extreme values pbest , global extreme values gbest cation between the species, so as to keep searching until it finds the place
} with the most food. The attributes of the individual are its location and
} While (whether meet the condition) search direction. The quantity of food corresponding to the location of
the individual can be understood as the ideal degree of the function

5
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

Fig. 7. Multi-robot searches for single source leak process in different scenarios.

6
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

Fig. 7. (continued).

7
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

Fig. 8. Multi-robot search path for single source leak process in different scenarios.

Table 3
Data of different numbers of robots in each scene in a single leak environment.
Num Scene 1 Scene 2 Scene 3 Scene 4

SR/% NI ST/s SR/% NI ST/s SR/% NI ST/s SR/% NI ST/s

2 98 25.61 54.9 10 31.28 62.60 80 23.86 48.68 25 20.12 40.13

3 99 20.63 42.85 92 29.20 60.16 93 21.55 40.57 90 20.10 37.40
4 100 19.54 39.18 93 27.19 54.59 95 18.8 36.87 97 18.8 36.96
5 100 19.03 38.61 93 23.31 47.29 95 16.34 31.25 98 17.46 36.84
6 100 18.99 37.92 96 22.39 44.76 96 16.29 30.87 99 17.25 36.38
7 100 18.90 36.15 98 21.31 42.86 97 15.84 30.30 99 16.72 32.13

value at that location. At the beginning, place all the particles in the bird in the bird group as an example, and update its position through
entire range, and then adjust your search direction and search span formula (6):
according to the historical optimal location searched by the individual
xk+1 = xkj + vkj (6)
and the historical optimal location searched by the entire population, so j

as to find the optimal location is the optimal solution.

Take birds predation as an example to analyze the PSO: suppose where, xk+1
j is the updated position, which is obtained by adding the
there are m individuals in the group P, each individual has its own initial speed to the k-th position, where the speed update formula is formula (7)
position x and initial velocity v, first evaluate the amount of food Y at
vkj = ω × vk−j 1 + c1 × r1 × pbest − xj + c2 × r2 × gbest − xj (7)
each individual’s location Each position is taken as an independent
variable into the objective function f (x), and the amount of food at the
where, ω is the inertia factor, С1 and С2 are learning factors, r1 and r2
current position is obtained. Through the mutual communication be­
are constants between (0,1), pbest is the local best variable, and gbest is the
tween the individuals, we can know that the mi bird has the largest
global best variable. ω plays a great role in the convergence of the
amount of food, then the mi bird is recorded as pbest, and the remaining
particle swarm optimization algorithm. The larger the value, the
birds record the position of the largest food gbest in the past. Take the j-th

8
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

particle update is affected by three factors: inertia, self, and population.

The final direction of motion is the sum of the items.

2.3. Search strategy

In this paper, based on PSO algorithm, multi robots are used to locate
and detect gas leakage sources. First, the continuous search space is
discretized, and the entire detection process is divided into three stages:
gas search, gas tracking and gas source confirmation.

(1) Gas search

According to the characteristics of multi-body robots that are

searched by multiple robots at the same time, this paper uses a divergent
search method to search for leaked gas. The robots start from the initial
position and search scatteredly is shown in Fig. 6. The blue circles
represent the robots, and the yellow arrow indicates the direction of the
robot’s movement. The robot gradually moves forward, covering the
entire area as far as possible to find the diffused gas.

(2) Gas tracking

Fig. 9. Search success rate of multiple robot systems with different numbers of Multiple robots conduct a divergent search. When a robot detects a
robots in a single leak environment in various scenarios. leaking gas, the robot sends a signal to other robots. The remaining
robots gradually move closer to the robot that detects the first gas
detected based on particle swarm optimization. When all robots detect
the leaked gas, the robot with the highest gas concentration is recorded
as the best robot rbest. The remaining robots gradually approach rbest and
cycle in turn, finally detecting the location with the highest gas
concentration.
In the process of robot transfer, according to the current fitness of
each robot individual, that is, the detected concentration value, find the
current best fitness of the individual and the best fitness of the group,
through the principle of PSO, according to formula (7) The speed is
updated, and the boundary speed is processed at the same time, and
finally the update is completed according to formula (6). The pseudo
code is shown in Table 1.

(3) Gas source confirmation.

The multi-robot performs multiple iterations in the stage of tracking

the gas source. When a certain robot reaches the optimal solution, the
robot stops moving and waits for other robots to approach the gas source
for confirmation. When two robots have the same measured concen­
tration and the same position. Then the gas source is confirmed
successfully.

(4) When detecting an unknown number of leak sources, after a gas

Fig. 10. Searching time of multiple robot systems with different numbers of
robots in a single leak environment in various scenarios.
source test is completed, multiple robots start a divergent search
from the current position, and in this cycle, complete all leak
source tests.
stronger the global search ability and the weaker the local search ability;
conversely, the local search ability is enhanced while the global search
3. Results and discussion
ability is weakened. The learning factors С1 and С2 are used to adjust
the weight of the particle’s own experience and group social experience
3.1. The search result
in its movement. vmax is the maximum speed to ensure the movement of
particles within a certain range and prevent search divergence (Eberhart
The particle swarm algorithm is used to coordinate the robot control
and Shi et al., 2001).
to realize the natural gas leakage source detection simulation process of
The first term of formula (7) is the memory term, which represents
the offshore platform. First, the number of robots is set. The robots start
the influence of the speed and direction of the last time; the second term
from their respective starting positions. It is assumed that the robot’s
is the self-cognition term, which is a vector from the current point to the
moving distance per second is the side length of a single grid. The po­
best point of the particle itself, indicating the source of the particle’s
sition information can be exchanged with each other to prevent colli­
action The third part is the group cognition item, which is a vector from
sions. The maximum number of iterations NI = 200. Set the initialization
the current point to the best point of the population, reflecting the
parameters as shown in Table 2, and simulate the location of the leak
collaboration and knowledge sharing between particles. Formula (6)
source by multi-robot collaboration.
and formula (7) can be represented by Fig. 5. The direction of the
Taking three robots cooperatively searching for a single gas source as

9
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

Fig. 11. The process of the Multi-robot searching for two leak sources in scene 1.

Fig. 12. The search path of the Multi-robot searching for two leak sources in Fig. 13. Success rate of different number robots in searching for two leak
scene 1. sources in scene 1.

an example, Fig. 7 shows the process of multi-robots cooperatively number of particles (number of robots) is m, the number of iterations is
searching for the gas source position in different scenarios. Because the n, and the complexity of the fitness function (the time it takes for a single
robots have the number and position of the gas source, when a certain robot to detect gas) is f, the overall running time of the algorithm is m ×
gas source is successfully detected, the best robot stays at the gas source, n × f. Since the time taken by a single robot to detect gas remains the
and other robots will continue to search for other air sources. Fig. 8 same, that is, the fitness function complexity f remains unchanged, each
shows that the multi-robots successfully locate a single air source and iteration requires fitness calculations, individual, individual, group
continue to search for the path they have taken in the four scenarios. optimal state selection and speed, location update, so the complexity of
This paper analyzed the computational complexity of the algorithm. the algorithm is:
Since the main input of PSO is the number of particles, the number of
(m, n) = O(n) * (O(m) + O(1) + O(m) + O(2m) + O(m)) ≈ O(m * n)
iterations, and the complexity of the fitness function, assuming that the

10
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

Table 4 3.2. The influence of the number of robots on search results

Data searched by robots in various scenes under different number of leak
sources. In order to explore the impact of the change in the number of robots
Num One source Two sources on the search success rate (SR/%), search time (ST/s), and number of
SR/% NI ST/s SR/% NI ST/s
iterations (NI) under different scenarios of a single leak source, we set
the minimum number of robots to 2 The maximum number is 7, and 100
2 98 25.61 54.9 6 45.11 91.68
search tests are performed on multiple robots under different numbers in
3 99 20.63 42.85 55 39.86 73.85
4 100 19.54 39.18 68 36.52 71.29 different scenarios. The search success rate, search time, and the average
5 100 19.03 38.61 82 34.32 65.12 number of iterations are shown in Table 3.
6 100 18.99 37.92 90 33.64 64.23 It can be seen from Fig. 9 that when the number of robots is two, the
7 100 18.90 36.15 94 33.01 62.68 detection rate of multi-robots to the gas source in the obstacle area is
low. When the number of robots is greater than or equal to 3, the
detection rate of the air source by the robot in the scene used is main­
tained above 90%. At the same time, the success rate of the robot suc­
cessfully detecting the air source position increases with the increase of
the number of robots. Because the arrangement of obstacles in different
platform scenarios will have a certain impact on the diffusion of leaked
gas and the movement of robots, the detection rate of multi-robots in
barrier-free scenarios is greater than that of obstacles, and multi-robots
in different scenarios. The detection rate is also different. According to
Fig. 10, as the number of robots increases, the time it takes to success­
fully detect the air source decreases. At the same time, the time it takes
for the robot to successfully locate the air source is different in different
scenarios. In the barrier-free environment of Scene1, the time it takes for
the robot to successfully locate the odor source is not the longest or the
shortest. So, the existence of obstacles is not completely detrimental to
the time it takes for multiple robots to locate the air source. When the
layout of the obstacles just limits the robot’s search area, and there is no
obstacle between the initial position of the robot and the linear distance
from the odor source, the robot can locate the source of the leak faster.

3.3. The influence of the number of gas sources on search results

Fig. 14. Success rate of different number robots in searching for one and two
leak sources in scene 1.
Take the multi-robot composed of three robots as an example. The
multi-robot searches and locates two odor sources in an obstacle-free
environment with two leak sources. The multi-robot search process
and search path are shown in Figs. 11and 12.
Multi-robots search under the two leak source environments in Scene
1. Starting from the initial position, the multi-robots successfully locate
the first air source position in 51-st second, and then continue to search,
successfully positioning at the 88th second. The second air source
location.
In order to explore the detection effect of different numbers of multi-
robots on two air sources, 100 tests were performed on different
numbers of multi-robots to calculate the average value. The positioning
effect of multiple robots on two air sources is shown in Fig. 13. When the
number of robots is 2, within 200 iterations, only 7% of the multi-robots
successfully locate two air sources, 85% of the times to locate one air
source, and 8% of the number of failed positioning. As the number of
robots increases from 2 to 7, the success rate of multi-robot positioning
for the two air sources gradually increases. When the number of robots is
greater than 3, the number of successful positioning of two air sources
reaches more than 50%, and the number of positioning failures gradu­
ally decreases.
The search data of different numbers of robots in different scenarios
Fig. 15. Searching time of different number robots in searching for one and under different leak source environments is shown in Table 4. Compare
two leak sources in scene 1. the influence of the different number of air sources on the successful
detection of air sources by multiple robots in the same scenario, as
shown in Fig. 14. In the two air source environments, the successful
It can be seen from the above formula that the computational
detection of two air sources marks a success. As the number of robots
complexity of the PSO is proportional to the number of particles and the
increases, the probability that the robot system successfully detects all
number of iterations. In this research, m and n are mainly considered for
gas sources gradually increases, and the detection success rate of the
analyzing search algorithms.
multi-robot system for a single gas source is greater than the detection
success rate for two gas sources under the same order of magnitude.
When the number of robots is far greater than the number of air sources,
the multi-robot has a good detection effect. In order to explore the

11
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

Fig. 16. The path of Multi-robot searching for two leak sources in different scenarios.

Table 5
Data of different numbers of robots in various scenes in two leak source environments.
Num Scene 1 Scene 2 Scene 3 Scene 4

NI ST/s NI ST/s NI ST/s NI ST/s

2 25.61 54.9 31.28 62.60 23.86 48.68 20.12 40.13

3 20.63 42.85 29.20 60.16 21.55 40.57 20.10 37.40
4 19.54 39.18 27.19 54.59 18.8 36.87 18.8 36.96
5 19.03 38.61 23.31 47.29 16.34 31.25 17.46 36.84
6 18.99 37.92 22.39 44.76 16.29 30.87 17.25 36.38
7 18.90 36.15 21.31 42.86 15.84 30.30 16.72 32.13

influence of the number of air sources on the time it takes for multiple respectively show the search path of multi-robots cooperatively
robots to successfully detect the air source in the same scenario, as searching for gas source locations in different scenarios. The robot
shown in Fig. 15, as the number of robots increases, the time it takes for searches from the starting position. When a certain gas source is suc­
the multi-robot system to successfully detect all air sources in the scene cessfully detected, the best robot stays at the gas source position and
decreases, and the detection the time taken by a single gas source is less performs concentration detection, and other robots will continue to
than the time used to detect two gas sources, which means that the more search for other gas sources.
the number of leaked gas sources in the platform, the more robots and The average values of parameters obtained in two leaks with various
more time it takes to have a good detection effect. scenarios after 100 tests are shown in Table 5. Due to the different
diffusion positions of the leaked gas in different scenarios and the
3.4. The influence of obstacle environment on search results different walking paths of the robots, Fig. 17 compares the time it takes
for the multi-robot to successfully detect the gas source with different
Simulating two gas sources leaking and spreading in different sce­ numbers of robots under the diffusion environment of two leak sources
narios, three robots perform collaborative search for two sources. Fig. 16 in different scenarios. Comparing the four curves, due to the different

12
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

divergent manner. When gas is retrieved, the robot with the largest
concentration is defined as the best robot, and the best robot guides
other robots towards moving in the direction of increasing concentra­
tion, the target positions of all robots are updated at the optimal robot
position according to formula (9). When the optimal robot position re­
mains stationary after multiple iterations, the iteration stops and the air
source position is the optimal robot position.
Xi+1 = Xbest + Li (9)

Fig. 17. Success rate of different number robots in searching for two leak
sources in different scene.

complexity of obstacles in each scene, the robot’s search success rate and
the time it takes to successfully search are different. Among them, the
robot has the highest search success rate in Scene 1 and the least time it
takes to successfully search for the gas source.

3.5. The influence of different algorithm on search results

This research choose the multi robot control strategies based on ACO Fig. 18. Multi-robot air source search success rate based on
(Meng et al., 2006; Zou et al., 2009) and CSA (Wang et al., 2016) are different algorithms.
selected respectively, and test the search efficiency of different algo­
rithms through comparative analysis. Based on the multi-robot search
strategy formed by the ACO algorithm, in the gas discovery phase, the
robot starts from the starting position and performs a random search.
When the gas is retrieved, the robot moves in the direction of the
maximum gas concentration, and the robot with the maximum con­
centration sends pheromone to other robots. Based on the probability
formula (8), other robots are attracted to move toward themselves, and
each movement step is a grid distance.

τj (t)α ηij (t)β

Pij (t) = ∑m α β
(8)
j=1 τj (t) ηij (t)

τij (t) = ci − cj

Pij (t) represents the probability that the best robot i attracts other robots
j at time t, and τj is pheromone information, which represents the con­
centration difference between robot i and robot j. ηij is heuristic infor­
mation, which represents the reciprocal of the distance between robot i
and robot j. α and β are parameters that determine the weightiness of
τj (t) and ηij (t).
Based on the multi-robot search strategy formed by CSA, in the gas
Fig. 19. Multi-robot air source search time based on different algorithms.
discovery phase, the robot starts from the starting position to search in a

Table 6
Data on the number of different robots based on different algorithms.
Num PSO ACO CSA

SR/% NI ST/s SR/% NI ST/s SR/% NI ST/s

3 99 20.63 42.85 71 66.02 135.18 81 43.12 88.60

4 100 19.54 39.18 80 61.93 119.31 85 40.08 77.13
5 100 19.03 38.61 88 58.82 118.40 93 37.64 74.66
6 100 18.99 37.92 93 53.42 108.16 93 34.56 66.20
7 100 18.90 36.15 98 49.68 94.32 96 30.55 63.64

13
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

Xi+1 is the position that the robot will reach in the next iteration, Xbest is gas source environments is longer than the time used in a single
the best robot position, L_i is a vector whose step size is distributed ac­ gas source environment.
cording to a Lévy distribution and angular direction is uniformly (3) The layout of obstacles on offshore platforms has a certain impact
distributed in the direction of increasing concentration of best robot. on multi-robot leak search. Different obstacle layouts may in­
We use three multi-robot search strategies based on different algo­ crease the time it takes for the robot to locate the leak source, and
rithms to conduct comparative experiments. First, take multi-robots may also reduce the time it takes for the robot to locate the leak
searching for a single leak source in an obstacle-free area as an source.
example, set the maximum number of iterations to 200, and each al­ (4) By comparing the multi-robot leak source search strategy based
gorithm runs independently 100 times and the average value of the on the three heuristic algorithms of PSO, ACO, and CSA in terms
required parameters as shown in Table 6. of search success rate, search, iteration number and search time,
Three meta-heuristic algorithms are used to analyze multiple robots. the results show that the improved PSO is superior than ACO and
The search success rates of the three algorithms are shown in Fig. 18. In CSA for control of multi-robot system in leakage source location.
the process of increasing the number of search robots from 3 to 7, the
search success rate increases as the number of robots increases. The According to this paper, the method of multi-robot collaborative
search success rate of the PSO algorithm is basically maintained at about search for natural gas leaking from offshore platforms is proposed and
100%, while the search success rates of the ACO and CSA are lower than analyzed, and the effectiveness of the search method is verified. This
that of the PSO. When the number of robots is smaller, the search success lays the foundation for our next step. We will establish an experimental
rates of the ACO and CSA algorithms are quite different from PSO. With platform, according to the proposed algorithm, based on the relevant
the increase in the number of robots, the search success rate of the three hardware and software equipment, simulate the collaborative search
algorithms gradually decreases. The ACO search success rate has process of multi robots for leaking gas in a real environment. During the
increased from 71% to 97%, which is the lowest search success rate experiment, when the robots successfully locate a certain leak source,
among the three algorithms. When the number of robots is 3, the success the robot transmits relevant information about the leak source to the
rate based on PSO has the largest increase compared to the success rate maintenance personnel for timely maintenance. At the same time, in the
based on ACO, which is 28%. When the number of robots is 7, the PSO- next work, we will continue to optimize the algorithm, analyze the ro­
based success rate is the smallest increase compared to the ACO-based bot’s motion mode, build a real simulation platform for analysis,
success rate, which is 2%. perform cost effectiveness accounting for multi robots, mainly including
Through 100 search experiments on multi robots of different orders mobile devices, communication equipment, multiple sensors and other
of magnitude, the time taken for the robot to successfully locate the air major devices, to optimize the use of costs and improve the practicality
source under each algorithm is averaged. The results are shown in of this method. Another future work is based on a variety of heuristic
Fig. 19. It can be seen from the figure that the time for the three algo­ algorithms to compare, using more precise probability models and other
rithms to successfully locate the gas source gradually decreases as the control algorithms to achieve multi-robot control to complete the
number of robots increases. Comparing these three algorithms, it can be detection of natural gas leaks on offshore platforms. In short, the method
seen that the time taken by the PSO algorithm to successfully locate the of multi-robot collaborative search for natural gas leakage from offshore
gas source is less than the time used by the ACO and CSA algorithms, and platforms based on PSO not only guarantees the normal operation and
stabilized at about 20s. According to the figure, it can also be seen that safe production of offshore platforms, but also contributes to the
the search time error of the ACO and the CSA to successfully locate the development of intelligence in the oil and gas field.
gas source is larger than that of the PSO. The reason is that the proba­
bility of the ACO algorithm attracts the wit of the robot and the existence Credit author statement
of the random step length of the robot movement in the CSA algorithm,
which leads to the increase of the average search time and the increase Tao Ma: Software, Writing- Original draft preparation, Investigation.
of the search time range. When the number of robots is 3, the search time Shuhai Liu: Conceptualization, Methodology, Reviewing, Funding
based on PSO has the largest increase compared to the search time based acquisition. Huaping Xiao: Software, Validation, Reviewing,
on ACO, which is 92.33s. When the number of robots is 7, the PSO-based Supervision.
search time has the smallest increase compared to the CSA-based search
time, which is 27.19s. Based on the analysis of Figs. 18 and 19, whether
it is from the search success rate or from the successful search time, the Declaration of competing interest
PSO positioning effect is significantly better than the ACO and CSA.
The authors declare that they have no known competing financial
4. Conclusion interests or personal relationships that could have appeared to influence
the work reported in this paper.
Different from the traditional method of using multi-sensor networks
to monitor natural gas leaks on offshore platforms, this paper proposes a Acknowledgement
multi-robot collaborative search method for natural gas leaks on
offshore platforms, dividing the leaked natural gas search process into This work was supported by Science Foundation of China University
three stages, gas discovery based on divergent walking, gas tracking and of Petroleum, Beijing (No. 2462020XKJS01).
gas source location based on PSO. Finally, we verifies the effectiveness
of the algorithm from the following aspects. References

(1) When multi robots locate the leak source based on PSO, an in­ Almeida, A.G., Vinnem, J.E., 2020. Major accident prevention illustrated by hydrocarbon
leak case studies: a comparison between Brazilian and Norwegian offshore
crease in the number of robots within a certain range can improve functional petroleum safety regulatory approaches. Saf. Sci. 121, 652–665. https://
the success rate of leak source location and detection efficiency. doi.org/10.1016/j.ssci.2019.08.028.
(2) Multi-robots have good location effects in a single leak source Bao, X., Sun, H., Iglesias, G., Wang, T., Shi, C., 2020. Signal denoising method for modal
analysis of an offshore platform. J. Loss Prev. Process. Ind. 63, 10400. https://doi.
environment and two leak source environments., the time taken
org/10.1016/j.jlp.2019.104000.
by the multi-robot to successfully locate all gas sources in the two Chen, X.X., Huang, J., 2019. Odor source localization algorithms on mobile robots: a
review and future outlook. Robot. Autonom. Syst. 112, 123–136. https://doi.org/
10.1016/j.robot.2018.11.014.

14
T. Ma et al. Journal of Natural Gas Science and Engineering 84 (2020) 103636

Consi, T.R., Atema, J., Goudey, C.A., Cha, J., Chryssostomidis, C., 1994. Auv guidance Marjovi, A., Marques, L., 2014. Optimal swarm formation for odor plume finding. IEEE
with chemical signals. In: Proceedings of IEEE Symposium on Autonomous Trans. Cybernet. 44 (12), 2302–2315. https://doi.org/10.1109/
Underwater Vehicle Technology. IEEE, Cambridge, MA, USA, pp. 450–455. https:// TCYB.2014.2306291.
doi.org/10.1109/AUV.1994.518659. AUV’94. Meng, Q.H., Li, J.C., Li, F., Zeng, M., 2006. Mobile robots odor localization with an
Das, P.K., Behera, H.S., Panigrahi, B.K., 2016. A hybridization of an improved particle improved ant colony algorithm. In: 2006 IEEE International Conference on Robotics
swarm optimization and gravitational search algorithm for multi-robot path and Biomimetics, Kunming IEEE, pp. 959–964. https://doi.org/10.1109/
planning. Swarm. Evol. Comput. 28, 14–28. https://doi.org/10.1016/j. ROBIO.2006.340358.
swevo.2015.10.011. Meng, Q.H., Yang, W.X., Wang, Y., Li, F., Zeng, M., 2012. Adapting an ant colony
Eberhart, R.C., Shi, Y., 2001. Particle swarm optimization: developments, applications metaphor for multi-robot chemical plume tracing. Sensors 12 (4), 4737–4763.
and resources. In: Proceedings of the 2001 Congress on Evolutionary Computation, https://doi.org/10.3390/s120404737.
vol. 1. IEEE, Seoul, South Korea, pp. 81–86. https://doi.org/10.1109/ Olsen, E., Næss, S., Høyland, S., 2015. Exploring relationships between organizational
CEC.2001.934374. factors and hydrocarbon leaks on offshore platform. Saf. Sci. 80, 301–309. https://
Farrel, J.A., Pang, S., Li, W., Arrieta, R., 2003. Chemical Plume Tracing Experimental doi.org/10.1016/j.ssci.2015.07.023.
Results with a REMUS AUV. Oceans 2003. Celebrating the Past ... Teaming toward Reeder, P.B., Ache, B.W., 1980. Chemotaxis in the Florida spiny lobster, panulirus argus.
the Future, vol. 2. IEEE Cat. No.03CH37492), San Diego, CA, USA, pp. 962–968. Anim. Behav. 28 (3), 831–839. https://doi.org/10.1016/S0003-3472(80)80143-6.
https://doi.org/10.1109/OCEANS.2003.178458. IEEE. Richart, V.R., Christian, D.O., Efraín, Q.P., Sam, M., 2016. A CFD-based approach for gas
Feng, Q., Cai, H., Li, F., Yang, Y., Chen, Z., 2018. Locating time-varying contaminant detectors allocation. J. Loss Prev. Process. Ind. 44, 633–641. https://doi.org/
sources in 3D indoor environments with three typical ventilation systems using a 10.1016/j.jlp.2016.03.004.
multi-robot active olfaction method. Build. Simul. 11, 597–611. https://doi.org/ Rozas, R., Morales, J., Vega, D., 1991. Artificial smell detection for robotic navigation.
10.1007/s12273-017-0424-6. In: Fifth International Conference on Advanced Robotics ’Robots in Unstructured
Feng, Q., Cai, H., Yang, Y., Xu, J., Jiang, M., Li, F., Li, X., Yan, C., 2020. An experimental Environments. IEEE, Pisa, Italy, pp. 1730–1733. https://doi.org/10.1109/
and numerical study on a multi-robot source localization method independent of ICAR.1991.240354.
airflow information in dynamic indoor environments. Sustain. Cities Soc. 53, Seo, J.K., Kim, D.C., Ha, Y.C., Kim, B.J., Paik, J.K., 2013. A methodology for determining
101897. https://doi.org/10.1016/j.scs.2019.101897. efficient gas detector locations on offshore installations. Ships Offshore Struct. 8 (5),
Hayes, A.T., Martinoli, A., Goodman, R.M., 2002. Distributed odor source localization. 524–535. https://doi.org/10.1080/17445302.2012.713219.
IEEE Sensor. J. 2 (3), 260–271. https://doi.org/10.1109/JSEN.2002.800682. Shallcross, D.C., 2013. Using concept maps to assess learning of safety case studies–The
Jain, U., Tiwari, R., Godfrey, W.W., 2019. Multiple odor source localization using Piper Alpha disaster. Educ. Chem. Eng. 8 (1), e1–e11. https://doi.org/10.1016/j.
diverse-PSO and group-based strategies in an unknown environment. J. Comput. Sci. ece.2013.02.001.
34, 33–37. https://doi.org/10.1016/j.jocs.2019.04.008. Shukla, A., Karki, H., 2016. A plication of robotics in onshore oil and gas industry-A
Khalili, A., Rastegarnia, A., Islam, M.K., Zhi, Y., 2013. A bio-inspired cooperative review partI. Robot. Autonom. Syst. 75, 490–507. https://doi.org/10.1016/j.
algorithm for distributed source localization with mobile nodes. In: 35th Annual robot.2015.09.012.
International Conference of the IEEE Engineering in Medicine and Biology Society Tirabassi, T., Tiesi, A., Buske, D., Vilhena, M.T., Moreira, D.M., 2009. Some
(EMBC). IEEE, Osaka, pp. 3515–3518. https://doi.org/10.1109/ characteristics of a plume from a point source based on analytical solution of the
EMBC.2013.6610300. two-dimensional advection–diffusion equation. Atmos. Environ. 43, 2221–2227.
Kim, J.K., Lee, D.H., Hong, Y.M., Ahn, S.H., Park, J.B., Seo, J.K., Kim, B.J., Pail, J.K., https://doi.org/10.1016/j.atmosenv.2009.01.020.
2016. Methods for determining the optimal arrangement of water deluge systems on Villarreal, B.L., Olague, G., Gordillo, J.L., 2016. Synthesis of odor tracking algorithms
offshore installations. Ocean Eng. 114, 236–249. https://doi.org/10.1016/j. with genetic programming. Neurocomputing 175, 1019–1032. https://doi.org/
oceaneng.2016.01.010. 10.1016/j.neucom.2015.09.108.
Kuwana, Y., Shimoyama, I., Sayama, Y., Miura, H., 1996. Synthesis of pheromone- Wang, W., Cao, M., Ma, S., Ren, C., Zhu, X., Lu, H., 2016. Multi-robot Odor Source Search
oriented emergent behavior of a silkworm moth. In: IEEE/RSJ International Based on Cuckoo Search Algorithm in Ventilated Indoor Environment. 2016 12th
Conference on Intelligent Robots and Systems. IROS ’96, vol. 3. IEEE, Osaka, Japan, World Congress on Intelligent Control and Automation (WCICA). IEEE, Guilin,
pp. 1722–1729. https://doi.org/10.1109/IROS.1996.569043. pp. 1496–1501. https://doi.org/10.1109/WCICA.2016.7578817.
Li, W., 2007. An iterative fuzzy segmentation algorithm for recognizing an odor source in Wei, T., Xu, M., 2016. An integral equation approach to the unsteady
near shore ocean environments. 2007 International Symposium on Computational convection–diffusion equations. Appl. Math. Comput. 274, 55–64. https://doi.org/
Intelligence in Robotics and Automation. Jacksonv. IEEE 101–106. https://doi.org/ 10.1016/j.amc.2015.10.084.
10.1109/CIRA.2007.382843. Wiedemann, T., Manss, C., Shutin, D., 2018. Multi-agent exploration of spatial dynamical
Lilienthal, A., Reimann, D., Zell, A., 2003. Gas Source Tracing with a Mobile Robot Using processes under sparsity constraints. Aut. Agents Multi-Agent Syst. 32, 134–162.
an Adapted Moth Strategy. Autonome Mobile Systeme 2003. Springer, Berlin, https://doi.org/10.1007/s10458-017-9375-7.
pp. 150–160. https://doi.org/10.1007/978-3-642-18986-9_16. Wiedemann, T., Shutin, D., Lilienthal, A.J., 2019. Model-based gas source localization
Liu, C., Wang, Y., Li, Y., Xu, M., 2018. Experimental study on new leak location methods strategy for a cooperative multi-robot system—a probabilistic approach and
for natural gas pipelines based on dynamic pressure waves. J. Nat. Gas Sci. Eng. 54, experimental validation incorporating physical knowledge and model uncertainties.
83–91. https://doi.org/10.1016/j.jngse.2018.03.023. Robot. Autonom. Syst. 118, 66–79. https://doi.org/10.1016/j.robot.2019.03.014.
Loutfi, A., Coradeschi, S., Karlsson, L., Broxvall, M., 2004. Putting olfaction into action: Yang, Y., Feng, Q., Cai, H., Xu, J., Li, F., Deng, Z., Yan, C., Li, X., 2019a. Experimental
using an electronic nose on a multi-sensing mobile robot. 2004 IEEE/RSJ study on three single-robot active olfaction algorithms for locating contaminant
International Conference on Intelligent Robots and Systems (IROS), Sendai. IEEE 1, sources in indoor environments with no strong air flow. Build. Environ. 155,
337–342. https://doi.org/10.1109/IROS.2004.1389374. 320–333. https://doi.org/10.1016/j.buildenv.2019.03.043.
Lytridis, C., Virk, G.S., Kadar, E.E., 2006. Search Performance of a Multi-Robot Team in Yang, Z., Abujaafar, K.M., Qu, Z., Wang, J., Nazir, S., Wan, C., 2019b. Use of evidential
Odor Source Localization. Climbing and Walking Robots. Springer, Berlin, reasoning for eliciting bayesian subjective probabilities in human reliability analysis:
pp. 809–816. https://doi.org/10.1007/3-540-26415-9_97. a maritime case. Ocean Eng. 186, 106095. https://doi.org/10.1016/j.
Marangoz, S., Amasyalı, M.F., Uslu, E., Çakmak, F., Altuntaş, N., Yavuz, S., 2019. More oceaneng.2019.05.077.
scalable solution for multi-robot–multi-target assignment problem. Robot. Autonom. Zhou, C., Zhang, B., Qiao, S., Mu, C., 2019. A stochastic programming approach for the
Syst. 113, 175–185. https://doi.org/10.1016/j.robot.2019.01.005. optimization of gas detector placement in offshore platforms. Ocean Eng. 187,
Marjovi, A., Marques, L., 2011. Multi-robot olfactory search in structured environments. 106233. https://doi.org/10.1016/j.oceaneng.2019.106223.
Robot. Autonom. Syst. 59, 867–881. https://doi.org/10.1016/j.robot.2011.07.010. Zou, Y., Luo, D., 2008. A modified ant colony algorithm used for multi-robot odor source
Marjovi, A., Marques, L., 2013. Swarm robotic plume tracking for intermittent and time- localization. In: International Conference on Intelligent Computing, vol. 5227.
variant odor dispersion. In: Mobile Robots (ECMR), 2013 European Conference on Springer, Berlin, pp. 502–509. https://doi.org/10.1007/978-3-540-85984-0_60.
Mobile Robots. IEEE, Barcelona, pp. 379–384. https://doi.org/10.1109/ Zou, Y., Luo, D., Chen, W., 2009. Swarm robotic odor source localization using ant
ECMR.2013.6698871. colony algorithm. In: 2009 IEEE International Conference on Control and
Automation, Christchurch, pp. 792–796 https://doi.org/ICCA.2009.5410516.

15