Fogarty et al [12] presents a low-cost microphone-based
sensor system for elder activity sensing. Battery powered
sensors are attached to pipes at critical locations in the household’s
water distribution system. These sensors collect audio
samples every 2 seconds which can be used to identify end
use activities. Although water pipes are good conductors of
sound, this also implies that they will conduct the ambient
noise well (e.g. sound of central air conditioning) which can
lead to erroneous detection of events. Since the sensors are
mounted to pipes, a plumber is not needed for installation.
However, the suitable placement of the sensors is essential for
the system to function correctly and professional installation
of the system will be required. However, proper placement
of the sensors leads to accurate detection of events including:
94% of shower usage; 95% of dishwasher usage; as well as 73
and 81% of bathroom and kitchen sink activity, respectively,
lasting 10 seconds or longer.
Larson et al [13] uses a centralised low-cost pressurebased
sensor for automatic disaggregation of water usage
events in ten households. The opening and closing of water
fixtures causes a pressure wave (i.e. surge/water hammer)
which propagates through the water distribution system which
is observed by the sensor. This pressure wave signature differs
depending on the valve type, location and the way in which
the valve is opened or closed. These pressure transients are
then classified into specific end uses as well as individual
fixtures. For example, if two identical toilets are located in
two separate bathrooms of a single household, each pressure
wave traverses a different path to the sensor, creating unique
signatures which can be used to differentiate between the
two fixtures. The system is able to classify the end uses of
water, the specific valve used during usage events and classify
events as hot or cold water events with accuracies greater than
90%. Additionally, the flow rates of individual fixtures were
estimated in four households and with three of the four had
error rates less than 8%. The fourth household had an error
rate of 22%which is believed to be as a result of incorrect
placement of the sensor.