Structural Health Monitoring using Wireless Sensor Networks - Mobile and Pervasive Computing in Construction - page 214

and biological detection and so on (Mainwaring et al ., 2002; Lo et al ., 2005;
Bokareva et al ., 2006), monitoring a civil structure is more difficult in many aspects.
For example, vibration data, collected using accelerometers, is one of the most
important types of data used in SHM. Vibration-based SHM requires sampling
frequency to be high enough to capture damage-sensitive information contained in
the measured signals. For robust estimation or due to the heavily damped nature of
most civil structures, sampling at hundreds of Hertz is quite common (Paek
et al ., 2005). This level of sampling frequency is rarely achieved in other WSN
applications. Besides, most of the SHM applications require synchronized sam-
pling. The synchronization error between any two sensor nodes should be no more
than 1ms to avoid any undesirable consequences to SHM algorithms (Nagayama
and Spencer, 2008). In environmental monitoring, sensing in different sensor
nodes is often implemented asynchronously.
Besides high frequency and strictly synchronized sensing, SHM requires that the
sampled data are reliably delivered, since most of the SHM algorithms cannot
tolerate data loss. It was found that more than 0.5% data loss causes noticeable
consequences on the damage detection results (Nagayama and Spencer, 2008).
Another important difference of SHM with other applications is associated with
SHM algorithms. Nearly all of the existing SHM algorithms are centralized and
require sampled data from multiple sensor nodes. In addition, the length of data
from each sensor node in most cases is larger than one thousand. Also, most of the
SHM algorithms are computationally intensive. The previous simple in-network
processing techniques used in environmental monitoring, such as mean/max/min
and so on, are rarely used in SHM. Obviously, implementing computational
intensive as well as resource-demanding SHM algorithms is more difficult than
implementing previous simple in-network processing algorithms.
Table 11.1 lists the main differences of SHM applications and other applications
of WSNs.
The comparably high resource requirement of SHM poses significant challenges
for resource-limited WSNs. Previously trivial tasks for a wire-based SHM system
can be very difficult, if not impossible, for a WSN.

Table 11.1 The main differences of SHM applications and other applications of WSNs

SHM application Other applications

Sampling High frequency (X00 times per

second)

Low frequency (X times per
second, minutes)

Synchronized sensing with
synchronization error G 1ms

Not necessarily synchronized

Data delivery Data loss rate G 0.5% Generally no such a strict

requirement

Processing algorithm Complicated Simple

1. SHM algorithms are centralized,
and require a bunch of data from
multiple sensor nodes

1. Algorithms are relatively
simple (mean/max/min) and
lightweight

2. SHM algorithms involves
complex signal-processing
algorithms and are
computationally intensive

2. Many algorithms are
distributed