Sunteți pe pagina 1din 8

2013 XXIV International Conference on

Information, Communication and Automation Technologies (ICAT)


October 30 November 01, 2013, Sarajevo, Bosnia and Herzegovina

Predictive Monitoring of Train Wagons Conditions


Using Wireless Network Technologies

Mirko Franceschinis, Francesco Mauro, Claudio Mario Rossi


Pastrone, Maurizio A. Spirito Railway & Off-Highway Business Unit
Pervasive Technologies (PerT) Area SKF Industrie S.p.A.
Istituto Superiore Mario Boella (ISMB) Airasca (Turin), Italy
Turin, Italy

AbstractPredictive monitoring of train wagons can allow to quantities like vibrations and temperature whose cross-analysis
anticipate possible malfunctioning due to wear and avoid allows to determine the state of the carriage and estimate the
potential accidents. In this paper some network architectures probability of faults within a certain timeline. The sensors are
adopting low-power wireless communication technologies are autonomous for energy provisioning, assuming they are
introduced. A performance comparison is provided based on ns-2 powered by specific harvester recovering energy from the
simulation results, suggesting that the combined use of WSN and environment. Sensor data locally collected by sensing nodes
WiFi in a hierarchical architecture is adequate for long trains are finally transmitted via radio communication to a gathering
with several coaches and a large number of sensing nodes. center. Several requirements in terms of bandwidth, radio
channel capacity, energy consumption, data security and
Keywordscondition-based monitoring; railway scenarios;
wireless sensors; ns-2 simulations
devices physical robustness have to be considered. In this
work, we focus our attention on communication aspects, which
will be presented and analyzed in major detail in section IV
I. INTRODUCTION AND MOTIVATION devoted to proposed system architectures. Before, section II
Passengers trains and, even more, freight cars cover long lists the main system requirements, in particular concerning
distances during the span of their railway life. National and data management and communication, and section III reviews
international railway legislation exists which imposes periodic the state of the art about wireless (particularly WSN-based)
checks of railway carriages conditions and the appropriate monitoring systems in railway scenarios. Simulation results
maintenance in railways workshops. Nonetheless the persistent involving the proposed architectures are illustrated and
monitoring of wagons conditions finalized to anticipate commented in section V. Section VI concludes the paper.
possible malfunctioning caused by wear and to avoid potential
accidents represents an added value. This concept is known as II. MAIN SYSTEM REQUIREMENTS
condition-based monitoring.
The sensing nodes integrated with bearings and positioned
The monitoring is made possible by the presence of sensors in axle-box must satisfy a set of physical restrictions in terms
installed on bearings positioned along the wheel-axle of the of sizes and deployment, as well as covers of adequate
several bogies and coaches composing a train. Bearings materials for device protection. Geometrical features also
deployed in axle-box collect acceleration samples which, concern carriage shape and train composition. Fig. 1 depicts the
opportunely correlated and fused, allow to derive real time shape of a real passenger coach, and the reported dimensions
information about the bogie state in terms of stability. Beside are used as inputs in the simulated scenario.
accelerometer values, temperature readings are of interest too
since overheating of bearings could compromise the correct
functioning and, definitively, the bogie solidity.
Bearings for railway bogie equipped with sensors already
exist which provide continuous time analogue signals.
However, the data communication is enabled by a wired
system and, similarly, the system energy requirements force the Fig. 1. A single carriage and, quoted, its dimensions in length.
bearings to be inline powered. The challenge, now, is instead to
realize a wireless communication system whose nodes are More important for communication system design are those
powered by energy harvesting solutions. A wireless system is requirements concerning temperature and acceleration data
desirable for costs reduction: installation and maintenance are sampling. Approximately, updated temperature readings should
simple and economic. be available about every minute while older samples than four
or five minutes are useless for application purposes.
Definitively, the system we have in mind can be described Acceleration data related to vibrations require to be collected
as follows. It consists of a number of sensors deployed in the much more frequently than temperature values since only long
carriage and able to capture, while the train is moving, physical

978-1-4799-0431-0/13/$31.00 2013 IEEE


sequences of samples allow to perform frequency analysis and A. Wireless technologies
then derive the bearings status (e.g., damage or wear). 1) IEEE 802.11
Another requirement is communication robustness against Throughout the years several proposals of the IEEE
possible interferences and disturbs caused by devices using working group on wireless local area networks have become
other wireless communication technologies in the same official IEEE 802.11 standards. They define radically different
frequency band. Data security and integrity are other relevant Physical (PHY) layers while keeping the Medium Access
issues, whose impact on the system design is in terms of Control (MAC) layer, based on carrier sense multiple access
augmented overhead for data cyphering and times for security (CSMA) approach, substantially unvaried. A list of IEEE
protocols execution. Further system primary requirements 802.11 standards is reported in TABLE I. along with the most
concern energy provisioning through harvesting solutions; relevant quantitative features: the frequency band and the bit-
however, the influence on communication aspects consists in rate, i.e. the channel transmission speed, which represents the
imaging energy saving techniques by switching off the radio of channel capacity of transferring bits (of a packet) once the node
sensing nodes only, while sink nodes are inline powered. has successfully gained the medium access. It is worth
stressing the difference between the bit-rate and the rate at
III. STATE OF THE ART which traffic is actually delivered, both measured in bit per
second (bps). The latter is just a small percentage of the bit-
A technology scouting addressed to real market products rate, and it is compromised by a number of factors like
witnesses that a commercial solution perfectly matching all the overhead due to packet headers, packet queuing delays,
monitoring system features we are interested in does not exist medium access latencies, collisions and possible
today. Closer topics to our target have been investigated within retransmissions. TABLE I. shows that IEEE 802.11n standard
research projects although these efforts have not been is able to guarantee a larger bit-rate in both the frequency bands
concretized yet in really tangible results. We briefly give at respectively 2.4 and 5 GHz.
evidence of those research results that mainly meet the most
relevant requirements and share many features with our target
TABLE I. IEEE 802.11 STANDARDS FEATURES
system.
IEEE Frequency Band Bit-rate
To the best of our knowledge, most of the academic papers Standard [GHz] [Mbps]
Year of release
dealing with wireless monitoring systems applied to railway 802.11 2.4 2 1997
scenarios focus on radio communications along the tracks [1, 802.11a 5 54 1999
2] while just few consider wireless transmission systems inside 802.11b 2.4 11 1999
trains. The interest is usually devoted to the communication 802.11g 2.4 54 2003
between a mobile element, the train, and a static one, the 802.11n 2.4 or 5 600 2009
ground. Communication issues involving passengers trains are
analyzed by a few papers, and mainly in connection with WiFi The IEEE does not test commercial equipment for checking
services. The only academic works concerning WSNs applied the compliance with the standards, thus the interoperability
on board the train for condition based monitoring purposes among devices might not be ensured. The non-profit WiFi
relate to freight trains, often composed of open wagons where Alliance [4] was constituted in 1999 to fill this gap and this
propagation issues loose of importance and are not as event has favored the term WiFi to be commonly used as a
challenging as in passenger trains [2]. Instead, [3] is relevant synonymous of IEEE 802.11x.
since it shares the same objective that has inspired our work,
that is, the use of WSN technology with sensor nodes deployed 2) IEEE 802.15.4
on board passenger trains to continuously monitor the bearings. Designed to enable low-cost short-range bidirectional
However, in that case the interest is mainly devoted to wave wireless communications, IEEE 802.15.4 is a standard aimed at
propagation characteristics around a train in order to derive industrial, scientific and medical (ISM) applications. It
hints for the antenna design and to the possibility of using specifies the PHY and the MAC layers for low-rate wireless
energy scavenging for power supply, while communication personal area networks (LR-WPAN). A first version of the
aspects are neglected. standard [5] was released in October 2003 while in September
2006 a second version was ratified [6]. TABLE II. summarizes
the main IEEE 802.15.4 features.
IV. PROPOSED ARCHITECTURES
After briefly introducing the main features of IEEE 802.11
TABLE II. IEEE 802.15.4 STANDARDS FEATURES
and IEEE 802.15.4, identified as the most appropriate radio
technologies for the mentioned application, we describe Frequency Band
Nation/Continent
Bit-rate
alternative communication architectures for the target railway [MHz] [kbps]
scenario. Traffic models and coherent different approaches to 868 Europe 20
915 USA 40
sensor sampling, data processing and information delivery are 2400 All (ISM band) 250
introduced and discussed as well. As already mentioned, we 780 China 250
implicitly assume that sensing nodes are constrained in terms 868/915 China 20 40
of energy being powered by some energy harvesting system, 950 Japan 20 100
which however is not a specific issue here.
B. Network architectures supported at the bottom tier where sensing nodes could need
Different system architectures and data management multi-hop routing to communicate with the sink node. In Fig.
schemes have been analyzed to evaluate how well each one fits 3, this is represented by red links which do not always connect
the requirements. A passenger train is composed of directly a sensing node with the sink node, as it happens
approximately ten/twelve coaches: we opt for 12 when running instead in Fig. 2.
simulations. We assume that each coach, as the one depicted in
Fig. 1, hosts 8 different sensing nodes, which must convey the
collected data towards a sink node positioned in the coach at
one side of the train (the locomotive).
Two main communication architectures are presented and
their suitability compared in this paper. The first one combines
IEEE 802.15.4 (WSN) and IEEE 802.11 (WiFi) to form a
Fig. 2. Single-hop WSN within a wagon.
hierarchical two-tier architecture. Each coach globally includes
9 nodes: 8 sensing nodes and 1 sink node. Sensing nodes
deliver sensor data packets to the sink node in the same coach,
leveraging on WSN communication. The sink nodes associated
with the 12 coaches transparently forward the received packets
to the final recipient, the sink node positioned in the
locomotive at one extremity of the train. The communication
among sink nodes is based on WiFi technology. Sink nodes
thus expose two radio interfaces: a WSN chip to interact with Fig. 3. Multi-hop WSN within a wagon.
the WSN section and a WiFi card to send/receive packets
finally destined to the system data collector. Summarizing, two 3) Flat mesh architecture
communication tiers characterize this system architecture: The flat mesh architecture is depicted in Fig. 4. Considering
First/Bottom tier, for WSN communication within each 12 coaches, the system is composed of 96 sensing nodes
coach, between sensing nodes and the sink node; organized in a large multi-hop IEEE 802.15.4 network. A
Second/Upper tier, involving WiFi communication unique sink node is positioned in the locomotive at the
among sink nodes. extremity of the train.

Two versions of the two-tier architecture are considered,


depending on the WSN network topology: mesh, if multiple
hops are necessary to make communicate a sensing node with
the sink node; star, if the sink node is in the radio coverage area
of every sensing node and a single-hop suffices.
The second architecture regards instead a flat mesh Fig. 4. Flat mesh architecture.
architecture, where only WSN communication technology is
considered. Differently from the double-tier architecture, a C. Data management schemes
unique sink node exists, deployed in the locomotive, instead of
12 sink nodes. The sensing nodes, whose total number is still The monitoring and communication system can be founded
equal to 12 * 8 = 96 (number of coaches times the number of on different data management approaches, which correspond to
sensing nodes per coach), communicate with each other equally different traffic generation models. Deciding which one
through multi-hop paths dynamically built in the mesh is the most adequate depends on many aspects, e.g. the nature
topology, pursuing the final aim of reaching the sink node. and characterization of the sensor reading. Basic principles are
described below.
1) Hierarchical two-tier architecture with single-hop WSN
In Fig. 2, each blue rectangle represents a coach, as Sample and transmit is the simplest approach. Each sensor
indicated by the stylized picture over the arrows. Small white reading, encapsulated in a packet, is immediately transmitted
circles are sensing nodes which send packets to the sink node, via radio to the system sink. To increase channel utilization
indicated as GW, standing for Gateway, due to its role. A star efficiency, consecutive samples can be stored as long as a
topology is considered for WSN network routing: the radio maximum-size data packets can be prepared. Sensing nodes are
coverage allows each sensing node to reach directly, without merely samplers and aggregators while they are not able to
any packet forwarding, the sink node. For this reason we call process data at all. Data processing operations are executed in a
this architectural configuration Hierarchical two-tier centralized fashion, locally at the sink node. This approach is
architecture with single-hop WSN. Sinks instead form an responsible for a periodic generation of a constant amount of
IEEE 802.11 ad hoc network. traffic so that, in the simulation environment, it has been
associated to a constant bit rate (CBR) traffic model.
2) Hierarchical two-tier architecture with multi-hop WSN
Process, fuse and transmit indicates the strategy of
Fig. 3 is pretty similar to Fig. 2. Indeed the same continuously sampling the sensor signal while, parallel,
communication technologies are used, but a mesh topology is
performing local processing of long data bursts by means of achieve specific scenario conditions like the induced topology.
filtering/fusion algorithms, so to send summarized contents. We discuss how we set up some of them in the simulations.
Transmission periodicity and amount of data processed before
a transmission are parameters to be set, even dynamically, The simulation duration is calculated long enough to
based on optimization purposes. Sensing nodes are required to provide different guarantees. First, at least 1000 data packets
perform onerous data processing and run complex fusion are transmitted by each source node, to enhance the statistic
algorithms taking advantage of significantly smaller amount of reliability of performance results. Secondly, it is checked that a
generated traffic. Due to limitation in paper size, our simulation transitory phase of network behavior represents a very small
scenarios only consider the sample and transmit approach and fraction of the whole simulation test and that ergodic
ignore the process, fuse and transmit strategy which is left for conditions hold during the steady state phase.
future in-depth analysis. The Free Space propagation model, which only depends on
the distance between nodes, is adopted. The Two Ray Ground
V. SIMULATION RESULTS propagation model is ignored since in the scenario conditions
induced by the train dimensions reported in Fig. 1 it exactly
The alternative communication architectures presented in coincides with the Free Space model: at 2.4 GHz, a frequency
the previous section have been compared via simulation band shared by IEEE 802.15.4 and IEEE 802.11, the distance
methodology. This section presents a short introduction to ns-2 between the sink node and any sensing node is smaller than the
[7] along with the simulation parameters of main interest for cross-over distance, a threshold in the propagation
our purposes, defines the metrics at the basis of the mathematical equation. The Shadowing propagation model is
performance comparison and finally illustrates the achieved instead too restrictive and setting the related parameters path
results. loss exponent and shadowing deviation as suggested by the
scenario environment characterization does not allow the nodes
A. Introduction to ns-2 to communicate at all.
Ns-2 [7] is an open source discrete event simulator targeted
Nodes transmission power, communication technology and
at networking research. It supports a very large set of protocols
receiver sensitivity are changed according to the scenario under
over both wired and wireless networks and includes the
test: transmission power is increased if a star topology is
implementation of a great variety and heterogeneity of models.
desired and decreased to force a mesh topology.
An object oriented script language, called OTcl, is used for
defining simulation conditions by setting as desired a list of Nodes deployment determines nodes positioning affecting
parameters which depend on the scenario. the network topology. We consider the railway scenario
described in section IV.B: a train composed of 12 coaches each
Ns-2 is still very popular, although originally designed
one hosting 8 sensing nodes, one for each bearing. A
during the early nineties, being a stable and trustworthy
hierarchical two-tier architecture requires a sink node per
simulator. As reminded elsewhere in the paper, one of the main
coach, thus resulting in 9 nodes for coach, 8 sensing nodes and
ns-2 drawbacks today is the difficulty of associating more than
the sink node. The network animator NAM can help by
just one communication interface with a single device, in
providing a 2D representation of the scenario. For instance,
contrast with real devices currently used like smartphones and
Fig. 5 represents a NAM screenshot of a single coach, where
with the IoT paradigm.
some flying packets can be observed beside the 9 nodes. The 8
sensing nodes are labeled 1 to 8 and are positioned in
B. Ns-2 simulation parameters correspondence of the bearings, along the two wagon axis. The
Ns-2 allows to set a wide set of simulation parameters, sink node, labeled 0, is located on the right side of the coach.
pertaining to various aspects. We mention the most relevant This deployment is kept the same for any other coach. For a
ones, according to the simulation scenarios of our interest: flat mesh architecture only the coach coincident with the
locomotive contains a total of 9 nodes, including the sink. Any
nodes deployment, the position being defined by three
other coach only hosts 8 sensing nodes, and no sink.
spatial coordinates
time duration of the simulation trial
models parameters for data packets generation
propagation model parameters
the communication technology, involving both PHY
and Link/MAC layers
node transmission power
device receiver sensitivity
frequency band adopted by the communication
technology
channel width in the selected frequency band

Some of these parameters are kept unvaried from one Fig. 5. NAM: the node deployment in a coach according to the hierarchical
two-tier architecture.
simulation to another. On the contrary others are tuned to
Ns-2 provides several traffic generators. The simplest one, D. Performance evaluation
and the only one considered in the simulations results In this section the simulation results achieved for the
presented in this paper, is the CBR traffic model, able to different system architectures and scenarios described in
simulate a source of packets periodically transmitted. It is section IV are presented. CBR traffic model is used and the
based on two input parameters whose combined tuning allows only parameter kept unvaried in all the simulations is the
to set out the traffic load as desired: the transmission interval packet size, selected equal to 96 bytes in order to avoid
(TI), i.e., the period between two distinct consecutive packets fragmentation. Hence the resulting offered traffic only depends
generated by the same sensing node, and the packet size. on the TI value. When running simulations, larger TI values
corresponding to lower traffic loads were initially set while
C. Performance metrics shorter transmission intervals and higher traffic loads were
Simulation trials have been carried out to evaluate network gradually chosen until overload conditions were obtained and
performance, discuss adequacy of each solution highlighting the actual system capacity was identified. When the generation
pros and cons, finally identify and select the most appropriate rate of new packets by source nodes exceeds a certain
network topology and overall architecture able to meet the threshold, buffers occupation locally at nodes tends to rapidly
requirements. Performance metrics taken into consideration increase until packet discards and losses consequently follow.
concern latency, number of transmissions and packet loss Roughly speaking, system instability conditions are reached.
probability. The respective definitions are better detailed in the
1) Hierarchical two-tier architecture with single-hop WSN
following.
As mentioned in section V.A the concurrent simulation of
The latency is intended on a per packet basis. The latency two (or more) different technologies is not possible in ns-2,
of a packet is defined as the time period elapsed from the which does not conceive multi-radio devices. The trick to
instant when the packet is generated at the application layer of elude the problem is based on the idea to decompose the
the source node to the instant when the same packet is analysis, by simulating earlier the IEEE 802.15.4 section of the
delivered to the application layer of the final recipient node. In network and, subsequently, the Wi-Fi part which, in some way,
ns-2, these two time instants are associated with two specific receives the output of the IEEE 802.15.4 simulation as its own
events, marked as AGT (agent) and logged in the trace file input.
available at the end of the simulation, when an enough large
At the bottom tier, an IEEE 802.15.4 network rules a single
number of packets will be collected. A pdf (probability
coach. The simulations have been carried out with the standard
distribution function) of latencies can then be built by suitably
parameters set in ns-2 for the IEEE 802.15.4 communication
dividing into disjointed sub-intervals (bins) the range of
technology. The node transmission power has been set to make
experienced latencies. Statistical moments like mean value
each sensing node in the sink node radio coverage thus
(first order) and variance (correlated with the second order) can
enabling a star topology. Any sensing node is representative for
be calculated.
all the other network nodes in terms of average performance.
Similarly to the latency, the number of transmissions is a
If TI is set equal to 0.4 seconds the aggregate network
metric associated with a single packet. It is defined as the total
packet loss probability is negligible: 0.001. Very rare loss
number of transmissions a packet has experienced from the
events are reasonably caused by some queues overflow or
first transmission by the original source node to the final
temporary node overload, such that some packet is dropped or
delivery to the sink node. In the most general scenario, a multi-
not received. Fig. 6 and Fig. 7 respectively display the pdf of
hop network where packet losses can occur on a link basis, a
latency and number of transmissions, aggregated on all sensing
new transmission can happen due to two different reasons. If
nodes. The latency is very low although extremely rare samples
the MAC layer supports link-based acknowledgments and
up to around 3000 milliseconds occur, but are cut in the plot to
consequent retransmissions, as it happens for IEEE 802.15.4, a
better display the part of the curve where the 99.9% of samples
transmission can be triggered by a Time-Out expiration.
are concentrated. The number of transmissions is more
However, even the forwarding of a packet by an intermediate
frequently equal to one, rare retransmissions events are
node along the route from source to sink node represents a new
coherent with the negligible but not null packet loss
packet transmission.
probability.
The third performance metric, packet loss probability, is
Fig. 8 is a latency curve with the same semantics as Fig. 6
defined, for each end-to-end pair of source-destination (the
but specifically referring to sensing node 1, the furthest from
latter is always the sink node), as the ratio between the number
the sink node (see Fig. 5). The plot shows that retransmissions
of data packets originally generated by the source node but
occasionally occur causing longer latencies. Nonetheless, as
never delivered to the sink node and the total number of data
expected for a star topology, the performance of a sensing node
packets originally generated by the source node. Differently
is in line with the network aggregate performance. The same
from the two previous metrics, just a single probability value
holds for the number of transmissions, not reported here.
can be computed for each sensing node at the end of a
simulation. Aggregation is clearly possible, thus resulting in
network-based packet loss probability with respect to the many
per node-based packet loss probabilities.
Network: Latency for received packets node. Concerning the number of transmissions, Fig. 10
0.5
witnesses a decreasing rate following a negative exponential
0.45
fall where values from 1 to 5 are remarkable.
0.4
Sensing node 1, the most distant from the sink node,
0.35
experiences high latency values, quite commonly reaching 2
0.3 seconds and sometimes overpassing the barrier of 4 seconds.
The related graph is omitted since quite specular to the
pdf

0.25

0.2
correspondent network aggregate. Similar comments regard the
number of transmissions.
0.15
Network: Latency for received packets
0.1

0.05 0.6

0
0 5 10 15 20 25 30 35 40 45 50
0.5
Latency (ms)

0.4
Fig. 6. Star topology TI = 0.4 s Aggregated latency.

pdf
0.3
Network: Transmissions (including the first) for received packets
1
0.2
0.9

0.8
0.1

0.7
0
0.6 0 1000 2000 3000 4000 5000 6000
Latency (ms)
pdf

0.5

0.4
Fig. 9. Star topology TI = 0.1 s Aggregated latency.
0.3
Network: Transmissions (including the first) for received packets
0.2 1

0.1 0.9

0 0.8
1 2 3 4 5
Transmissions 0.7

0.6
Fig. 7. Star topology TI = 0.4 s Aggregated number of transmissions.
pdf

0.5

Node 1: Latency for received packets 0.4


0.35
0.3

0.3 0.2

0.1
0.25
0
1 2 3 4 5 6 7 8 9
0.2 Transmissions
pdf

0.15
Fig. 10. Star topology TI = 0.1 s Aggregated number of transmissions.
0.1
2) Hierarchical two-tier architecture with multi-hop WSN
0.05 A mesh topology inside a coach is forced when not all of
the sensing nodes lie in the radio coverage range of the sink
0
0 10 20 30 40 50 60 node. In ns-2, this scenario can be easily achieved by setting
Latency (ms) the transmission power smaller than the threshold value which
guarantees a star topology. The communication among sensing
Fig. 8. Star topology TI = 0.4 s Node 1 latency. nodes and the sink deployed in the same coach exploits multi-
hop routing protocols like AODV [8], whose implementation is
Reducing TI to 0.1 seconds brings to a proportional made available in ns-2. It is worth noting that, in presence of a
increase of the traffic load. The tangible effect is an aggregate mesh topology, the number of transmissions experienced by a
packet loss probability reaching the impacting value of 0.454. packet results from two distinct events: retransmissions driven
Fig. 9 shows that, not so infrequently, some latency values are by link losses or packet discard at queues; and, packet
extremely high, comprised between 1 and 3 seconds, forwarding at intermediate relay nodes in accordance with
sometimes even larger. Very often, this is the result of long multi-hop routing.
queues and consequently long waiting/service time locally at a
Similarly to what happens with the star topology scenario, 1, relate to received packets only, in this scenario just a
the traffic load in correspondence of TI equal to 0.4 seconds is minority. The performances strictly depend on the node
not high enough to overload the system. The average packet position. Packets generated by sensing node 1 experience 4 to 6
loss probability is again negligible, specifically equal to 0.003. transmissions, responsible for the particular shape of the
As reported in Fig. 11 the latency samples are mostly latency pdf, resembling a uniform distribution on specific short
concentrated in the interval [0;50] milliseconds and just a ranges of values, probably due to network mechanisms that
minor part of them is comprised between 25 and 50 make periodical some impacting events (see Fig. 15).
milliseconds. However some isolated sample suffers from
Network: Latency for received packets
latencies even close to 500 milliseconds. Fig. 12 shows that
almost half of the packets needs just one transmission, while
one or two additional transmissions occur on the average for 0.25
the remaining 50% of packets.
0.2
Network: Latency for received packets
0.7

pdf
0.15
0.6

0.5 0.1

0.4 0.05
pdf

0.3
0
0 5 10 15
Latency (ms)
0.2

0.1 Fig. 13. Mesh topology TI = 0.1 s Aggregated latency.

0 Network: Transmissions (including the first) for received packets


0 50 100 150 200 250 300 350 400 450 500 1
Latency (ms)
0.9

Fig. 11. Mesh topology TI = 0.4 s Aggregated latency. 0.8

0.7
Network: Transmissions (including the first) for received packets
1 0.6
pdf

0.9 0.5

0.8 0.4

0.7 0.3

0.6 0.2
pdf

0.5 0.1

0.4 0
1 2 3 4 5 6 7 8 9
0.3 Transmissions

0.2
Fig. 14. Mesh topology TI = 0.1 s Aggregated number of transmissions.
0.1

0 Node 1: Latency for received packets


1 2 3 4 5 0.18
Transmissions
0.16

Fig. 12. Mesh topology TI = 0.4 s Aggregated number of transmissions. 0.14

Sensing node 1 moves away from the network aggregate 0.12

behavior if the number of transmissions is considered: it is 0.1


pdf

deterministically equal to 3. In terms of latency, samples 0.08


distribution reminds the aggregate one.
0.06
A correct interpretation of Fig. 13 and Fig. 14, referring to
TI set equal to 0.1 seconds, cannot disregard the estimated 0.04

value of network packet loss probability, 0.748, an 0.02

unacceptable performance result associated with long queues, 0


dropped packets, long delays, system overloaded. Latency 0 50 100 150 200 250 300 350 400 450 500
Latency (ms)
values, concentrated in the narrow [0;15] milliseconds range,
and number of transmissions, almost deterministically equal to
Fig. 15. Mesh topology TI = 0.1 s Node 1 latency.
TABLE III. allows a comparison between star and mesh TABLE IV. FLAT MESH ARCHITECTURE PERFORMANCE RESULTS
topologies. Average refers to all the sensing nodes, Min/ Aggregate Packet Loss Probability
Max to the node with the best/worst performance. When TI is TI [sec]
Min Max Average
about 0.4-0.5 milliseconds a stable and reliable behavior is 120 0.004 1 0.985
assured by both the topologies. As TI decreases the star 60 0.860 1 0.981
topology appears more reliable: 0.454 vs 0.748 is the network 30 0.029 1 0.965
packet loss probability in the two scenarios when TI = 0.1 VI. CONCLUSIONS AND FUTURE WORK
seconds. As expected, the star topology works better; but the
performance gap is slight. Finally, the break-down input traffic The simulation results showed that only a hierarchical
achieved by simulation is just indicative, due to the wide set of architecture integrating WSN and WiFi is able to manage a
uncertainties: relative values are significant, absolute values are monitoring system consisting of tens or hundreds of sensing
not. nodes, i.e., a train composed of ten coaches or more. A flat
mesh architecture would soon result in network collapse since
the traffic load managed by relay sensing nodes is not uniform
TABLE III. STAR AND MESH TOPOLOGIES: A COMPARISON and varies super-linearly from farthest to closest nodes (w.r.t.
Aggregate Packet Loss Probability the sink node). Queues are quickly full of packets, very long
TI
[sec]
Star Mesh delays follow and network resources are inefficiently used due
Min Max Average Min Max Average to uselessly performed work (packets forwarded for a section
0.45 0 0 0 0 0 0 of the path before being dropped).
0.4 0.001 0.003 0.001 0 0.017 0.003
0.3 0.007 0.057 0.039 0 0.933 0.239 This conclusion holds as long as the sample and transmit
0.2 0.150 0.203 0.180 0 0.092 0.012 approach is adopted, which is responsible for high traffic load
0.1 0.365 0.522 0.454 0.001 0.999 0.748 generation if sensors are sampled with high frequency. The
process, fuse and transmit approach could be successful for
3) Hierarchical two-tier architecture: WiFi two basic reasons: first, it can strongly reduce the amount of
The upper tier of the two-tier architecture conceives WiFi data a sensing node sends on the channel; second, a valuable
communication, configured in ad hoc mode. The 12 sinks, one energy saving can derive from the possibility of disabling the
per coach, forward the packets received from sensing nodes radio if data transmissions are less frequent. Conversely, the
belonging to the same coach. Ns-2 provides a WirelessPhyExt drawback is that it could require very powerful but energy
library to ensure a correct WiFi implementation. Sink nodes are consuming processing units. This kind of investigation is the
arranged in a linear topology, equally spaced, and cooperate to first in the list for future work. Ns-2 provides more complex
route packets to the sink node positioned in the locomotive at traffic models than CBR which can be utilized to simulate a
one side of the train. data management coherent with process, fuse and transmit,
e.g. the Exponential ON/OFF model characterized by the
Our goal is to check which technology, IEEE 802.15.4 in alternation of silence (OFF) and activity (ON) periods
the single coach scenario or IEEE 802.11 in the inter-coaches according to negative exponential random variables.
scenario, constitutes the network bottleneck. Considering that Transmissions during an ON period are still in CBR mode.
each WSN sink generates in the WiFi network the equivalent
traffic of 8 sensing nodes, and reminding WSN results, WiFi REFERENCES
simulations have been carried out by varying TI from 450/8 =
[1] T. Ito, N. Kita, W. Yamada, M.C. Tseng, Y. Sagawa, M. Ogasawara, M.
56.25 ms to 100/8 = 12.5 ms. For sake of simplicity CBR Nakatsugawa and T. Sugiyama, Study of propagation model and fading
traffic is used in the WiFi scenario, a reasonable approximation characteristics for wireless relay system between long-haul train cars,
considering the many other scenarios uncertainties although the 5th European Conference on Antennas and Propagation (EUCAP),
assumption is not fully realistic. Rome, 2011.
[2] J. Kim, K.-S. Lee, J.-G. Oh, A Study on the Wireless Onboard
As expected, the fundamental result is that WiFi is not the Monitoring System for Railroad Vehicle Axle Bearings Using the SAW
system bottleneck, being able to drain the aggregate WSN Sensor, S-CUBE 2010.
traffic load without no packet losses or discards. [3] M. Grudn, A. Westman, J. Platbardis, P. Hallbjrner, A. Rydberg,
Reliability Experiments for Wireless Sensor Networks in Train
4) Flat mesh architecture Environment, Proceedings of the 2nd European Wireless Technology
The flat mesh architecture is composed of only one sink (in Conference
the locomotive) and 12*8 = 96 sensing nodes, where 12 is the [4] WiFi Alliance, http://www.wi-fi.org/
number of coaches and 8 is the number of sensing nodes per [5] IEEE Std. 802.15.4-2003, Wireless Medium Access Control (MAC)
coach. Independently of the specific coach belonging and and Physical Layer (PHY) Specifications for Low-Rate Wireless
relaying on multi-hop capability, each sensing node sends its Personal Area Networks (LR-WPANs), IEEE, 2003
packets to the unique sink node using IEEE 802.15.4. The [6] IEEE Std. 802.15.4-2006, Wireless Medium Access Control (MAC)
and Physical Layer (PHY) Specifications for Low-Rate Wireless
network performances are generally very bad, as reported in Personal Area Networks (LR-WPANs), IEEE, 2006
TABLE IV. Such topology is not applicable not only for the
[7] Ns-2 simulator, http://www.isi.edu/nsnam/ns/
most challenging accelerometers which require higher sample
[8] C. Perkins, E. Belding-Royer, S. Das, "Ad hoc On-Demand Distance
frequencies and larger amount of data, but also for the slow- Vector (AODV) Routing", Feb. 2003. http://www.ietf.org/internet-
varying temperature readings. drafts/draft-ietf-manet-aodv-13.txt

S-ar putea să vă placă și