Wireless Sensor Networks for

Civil Infrastructure Monitoring

ISBN 978-0-7277-6151-4

ICE Publishing: All rights reserved

http://dx.doi.org/10.1680/wsncim.61514.153

Chapter 21
Temporary earth restraining structure
(Singapore)

This case study describes the deployment of a prototype wireless sensor network (WSN) on
the temporary earth restraining structure (TERS) of a mass rapid transit (MRT) construction
site in Singapore, monitoring strain and environmental conditions. The primary goal of this
deployment is to validate the WSN design (i.e. wireless communication and validity of meas-
urements) and to provide an archive of high-resolution strain data (15 min intervals) for off-
line analysis.

21.1. Description
21.1.1 Asset
The asset comprises several I-beam struts on a TERS at an MRT excavation site in Singapore.

21.1.2 Background
The larger goals of this project are to apply wireless sensing to TERS monitoring in order to
reduce monitoring costs but maintain safety through dense, cheap measurements and integra-
tion of data into an evolving model of the construction (Wilkins et al., 2015). These initial
deployments are to evaluate the performance of wireless communication and data measure-
ments within the construction over a period of a number of months. For sensors, there are
several challenges and constraints that complicate their deployment:

n There is limited deployment access time, and a low priority for the installation of
sensor nodes.
n Deployments are ongoing with the progress of the construction.
n There is limited space available to deploy on the strut, and thus no capability for power
harvesting on sensor nodes.
n It is impractical to retrieve previously deployed instrumentation after excavation.
n The environment is harsh in terms of performing sensor calibration and deployment
validation.

21.1.3 Monitoring objectives

The prototype network has several goals, as follows:

n Strain measurements are to be made on-strut.

n Data are to be transmitted wirelessly at 15 min intervals, to target a 1 year lifetime (or
the length of the construction).
n The data must be able to integrate into a TERS model.

153

Downloaded by [ Monash University] on [26/04/19]. Copyright © ICE Publishing, all rights reserved.
 Wireless Sensor Networks for Civil Infrastructure Monitoring

21.2. Solution
Figure 21.1 shows the basic network architecture, where sensor data flow from the network
to a data sink, where such data are transmitted to a remote server by way of 3G and are
subsequently made available to suitable applications (such as a graphical user interface, or a
model).

Given the short (12-month) development time for the monitoring system, most of the hardware,
software and mechanical parts were off the shelf. The data sink and wireless sensor nodes were
designed to be resistant to the tropical environment conditions in Singapore, where tempera-
tures can rise to 35°C and 99% relative humidity above ground, and even more inside the
excavation (40°C or higher). The size of the nodes needed to be small to reduce the likelihood
of being damaged on site, and thus power-harvesting capabilities were unavailable.

The sensor node is based around the Zolertia Z1 platform, with a bespoke strain gauge board
(see Figure 21.2). It is backwards compatible with the ubiquitous Telos platform, but
includes larger amounts of program memory, and better access to hardware pins with which
to interface the selected strain sensor board. The platform is a combination of the MSP430
microcontroller and a 2.4 GHz CC2420 802.15.4 radio, and supports an external antenna.
The Z1 supports both popular open-source WSN operating systems (OSs), TinyOS and Con-
tiki OS, out of the box.

The static strain measurement was chosen to be resistive, instead of using more costly vibrat-
ing wire gauges. Because the strain in a strut is assumed to be axial, only one strain channel
is required per sensor. A custom printed circuit board was developed to support a single
Wheatstone bridge circuit in a quarter-bridge configuration, which could be read by a low-
power 16-bit analogue–to-digital converter (ADC). To save energy, half of the bridge was
configured to use 10 kΩ resistors. The bridge can be balanced using a manual potentiometer.
The low-power ADC was connected to the Z1 by way of the inter-integrated circuit (I2C)
digital interface. The strain measurement resolution is <1 με, with a measurement range of
±2500 με. Input for an external SHT15 digital temperature and humidity sensor was included,
so that it could be embedded in the node packaging.

Figure 21.1 Overview of the WSN for TERS monitoring

154

Downloaded by [ Monash University] on [26/04/19]. Copyright © ICE Publishing, all rights reserved.
 Temporary earth restraining structure (Singapore)

Figure 21.2 Wireless sensor network hardware used for TERS monitoring

The sensor nodes are packaged in an off-the-shelf IP65 aluminium enclosure

(115 mm × 65.5 mm × 50 mm, weighing 0.37 kg). The enclosure is machined to support
three external attachments: an externally mounted radio antenna (4.4 dBi Antenova Titanis),
a gland to support the strain gauge wires going into the node and a port to allow the tem-
perature/humidity sensor to be exposed to outside weather conditions without allowing water
ingress. Magnets were bolted on to the enclosure to allow the nodes to be placed over the
strain gauge during the deployments.

The application code is built on top of the Contiki OS (Dunkels et al., 2004). The Contiki OS
was chosen for its ease of use (existing C code was easily transferred over) and the availabil-
ity of off-the-shelf components for building a WSN. In particular, the Contiki OS has support
for low-power communication through ContikiMAC (Dunkels, 2011) and the collect appli-
cation (Dunkels, 2007) – equivalent to the collection tree protocol (CTP) in TinyOS (Gna-
wali et al., 2009) – which creates a low-power, collection-tree-based multi-hop network that
can adapt dynamically to changing network conditions.

Application software was built on top of this to record network information as well as to
acquire and store strain and environmental data for subsequent offline analysis. For this pilot,
the software was designed as ‘sense, store and send’; that is, in normal operation, the node
samples every 15 min, stores the data locally in serial flash memory and then transmits the
data over the multi-hop network to the data sink.

A micro-benchmarking methodology was used to estimate the lifetime of the sensor nodes
with a given battery capacity and sampling at a certain rate. The individual aspects that make
up a sensing cycle of 15 min were measured and aggregated, with the resulting numbers used
to determine a baseline for the node lifetime given a specific battery size. A maximum of
270 days’ operation was expected with 7800 mA h C-cell batteries. Power consumption can

155

Downloaded by [ Monash University] on [26/04/19]. Copyright © ICE Publishing, all rights reserved.
 Wireless Sensor Networks for Civil Infrastructure Monitoring

be further reduced by using model-based techniques for radio transmission suppression

(Gaura et al., 2013).

Raspberry Pi (model A+) was chosen as the platform for the data sink. It is an embedded
Linux platform that has excellent hardware and software support and an active development
community. A Telos node connected over a universal serial bus (USB) is used as a sink to
link the gateway to the sensor network; it pushes all messages received over the serial port. A
USB 3G modem is used to transmit WSN data to a remote server every hour. A USB Wi-Fi
dongle was also added for on-site debugging: the Pi runs a web server and acts as a Wi-Fi
access point (using HostAP), so a field technician can connect to the Wi-Fi network of the Pi
on-site, and look at up-to-date statistics of incoming data in a web browser. Power to the
USB connections is controlled in software, saving power by only switching them on when
they are needed.

As with the sensor nodes, data are always stored locally, meaning any network problems will
not affect the final datasets. In this set-up it was chosen to power solely by battery and,
through micro-benchmarking, it is estimated that, when using a 100 A h battery, the lifetime
of the device is just over one calendar month. This level of lifetime is reasonable for battery
replacement, but for future solutions it is planned to use solar power to remove the need for
frequent battery replacement.

21.3. Monitoring results

From February to April 2015 it was possible to carry out the deployment of the system on a
live excavation. The main goals were to validate the performance of the system in situ, in
terms of:

(a) measurement validity, when compared against vibrating wire and load measurements
being taken onsite as part of the construction (on the same struts)
(b) wireless performance, in terms of data yield and network stability.

The main concerns with resistive strain measurement were related to the ability to provide a
reliable, stable static strain measurement over long periods of time (several months) without
drift. Two sensors were deployed on every strut monitored in order to allow redundancy and
to understand more about the WSN behaviour over time.

Strain gauges and sensor nodes were deployed on a level 2 strut at the excavation, about
10 m below ground (see Figure 21.3). The gateway was deployed at ground level at the side
of the excavation, with the distance between the nodes and the gateway around 12 m. The
gauges were welded to the strut after preparation, alongside the contractor’s vibrating wire
system (wired acquisition), and before pre-loading. It was not required to deploy the node
enclosures directly over the strain gauges, and instead they could be placed on the lower
webbing of the strut.

This deployment lasted at least 70 days. After this period the gateway was moved to another
part of the construction site for new measurements, but the nodes could not be retrieved until
the end of the construction works.

156

Downloaded by [ Monash University] on [26/04/19]. Copyright © ICE Publishing, all rights reserved.
 Temporary earth restraining structure (Singapore)

Figure 21.3 Example deployment on a strut next to vibrating wire gauges

The two nodes had data yields of 81.6% and 80.9% over the deployment period. This was
affected by several construction-related problems. For example, the gateway’s antenna was
covered for a period, and this caused an outage, as did lack of access to change the server
battery. In general the data yield was very good (close to 100%) and the network was stable.
Most of the time, node 1 was able to connect directly to the data sink, and, for about 50% of
the time, node 2 sent data through node 1.

In CTP and the Contiki OS collect application, a node’s wireless stability can be estimated by
how often it sends ‘beacons’ to find other nodes within communication range. If the connec-
tion with its parent and children nodes is stable, it will stay at the maximum 3600 s (1 h)
beacon interval. For these two nodes, this happened for at least 95% of the time. At the sink,
the received signal strength was within expected bounds ( 75 dBi and 83 dBi average),
with the message reception ratio dropping off at around 90 dBi, which is the sensitivity
level of the radio.

Figure 21.4 shows the measurements taken from the two strain gauges over several months
(no thermal compensation). Initially the measurements matched very well, but after 2 weeks,
strain gauge no. 2 showed abnormalities, sometimes dropping to very low strain readings
that were not reflected by the other gauge. This eventually manifested itself in a permanent
drop in strain reading. The other gauge appeared not to have these problems, which were
attributed to weather effects, perhaps water ingress into the sensor node enclosure.

Figure 21.5 shows the strain measurements converted into load (using the known dimensions
of the strut and the elastic modulus), and compared with the output of a load cell deployed on
the strut.

157

Downloaded by [ Monash University] on [26/04/19]. Copyright © ICE Publishing, all rights reserved.
 Wireless Sensor Networks for Civil Infrastructure Monitoring

Figure 21.4 Two wireless sensor strain gauge data traces

158

Downloaded by [ Monash University] on [26/04/19]. Copyright © ICE Publishing, all rights reserved.
 Temporary earth restraining structure (Singapore)

Figure 21.5 Estimated load plotted against load cell measurements

159

Downloaded by [ Monash University] on [26/04/19]. Copyright © ICE Publishing, all rights reserved.
 Wireless Sensor Networks for Civil Infrastructure Monitoring

In Figure 21.5, the top graph shows the vibrating wire estimated load plotted against the load
cell output, and the bottom graph shows the welded strain gauge estimated load against the
load cell output. The load cell clearly shows a diurnal pattern that is in synchronization with
the welded gauge strain measurements, although the welded strain has a larger magnitude.
These patterns are most likely a result of the effect of temperature variation. The vibrating
wire measurements show little or no variation with temperature, and have likely been pre-
compensated for thermal output.

Acknowledgements
This research was supported by the National Research Foundation Singapore through the
Singapore MIT Alliance for Research and Technology’s Center for Environmental Sensing
and Modeling interdisciplinary research program. Fieldwork was enabled through a collab-
orative agreement between Smart and Singapore’s Land Transport Authority.

REFERENCES
Dunkels A (2007) Rime – a lightweight layered communication stack for sensor networks. In
Proceedings of the European Conference on Wireless Sensor Networks (EWSN), Delft, The
Netherlands (Langendoenk K and Voigt T (eds)). Springer, Berlin/Heidelberg, Germany.
Dunkels A (2011) The ContikiMAC Radio Duty Cycling Protocol. Swedish Institute of Computer
Science (SICS), Stockholm, Sweden, Technical Report T2011:13.
Dunkels A, Gronvall B and Voigt T (2004) Contiki – a lightweight and flexible operating system
for tiny networked sensors. In Proceedings of the 29th Annual IEEE International Conference
on Local Computer Networks (Jha S, Hassanein H, Bulusu N et al. (eds)). IEEE Computer
Society, Washington, DC, USA, pp. 455–462.
Gaura E, Brusey J, Allen M et al. (2013) Edge mining the internet of things. IEEE Sensors Jour-
nal 13(10): 3816–3825.
Gnawali O, Fonseca R, Jamieson K, Moss D and Levis P (2009) Collection tree protocol. Pro-
ceedings of the 7th ACM Conference on Embedded Networked Sensor Systems (SenSys ’09),
Berkeley, California, CA, USA, pp. 1–14.
Wilkins R, Gaura E, Allen M et al. (2015) Restructure: a wireless sensor network for monitoring
temporary earth retaining systems. Proceedings of the 6th ACM Workshop on Real World Wire-
less Sensor Networks, Seoul, Republic of Korea, pp. 37–40.

160

Downloaded by [ Monash University] on [26/04/19]. Copyright © ICE Publishing, all rights reserved.