Database Reference
In-Depth Information
March 22, 2006, Setup 1, Sensor 1: Leak vs. No-Leak Power Spectrums
Leak
-135
Normal
-140
-145
-150
-155
-160
-165
-170
-175
-180
0
50
100
150
200
250
300
350
400
450
Frequency (Hz)
Figure 11.1
(See color insert following page 224.) A leak shows up as ad-
ditional energy in characteristic frequency bands.
In most applications, processing and analyzing these streaming sensor sam-
ples requires nontrivial event-stream and signal-oriented analysis. In many
cases, signal processing is application-specific and hence requires some amount
of user-defined code. For example, in a pipeline rupture detection application
with which we are familiar, the incoming streams of pressure data from sen-
sors on water pipelines, sampled at 600 Hz, are fed into a frequency transform.
Ruptures are detected by comparing the energy in certain frequency bands to
look for a peak. This is shown in Figure 11.1. Operations like “peak extrac-
tion” require user-defined code and are essentially impossible to implement in
simple languages like StreamSQL.
As another example, we have been working with biologists in Colorado to
build sensor systems that acoustically detect and localize marmots (a kind
of rodent endemic to the western United States) by listening for their loud,
chirpy calls. These systems consist of several four-microphone arrays that are
positioned around an area known to have marmots. Each microphone array
“listens” for sound frequencies that are characteristic of marmots and, when
it detects such signals, performs a “direction of arrival” analysis to determine
the bearing of the marmot. By finding the point where the direction of arrival
