FIGURE 2: Schools are using cameras paired with machine learning to identify and flag
individuals carrying weapons on school grounds. With hundreds of cameras per school,
this would be impossible without edge computing.
32 I ICT TODAY
feeding them, the result will be
an avalanche of computing
happening at the edge. The
sensor, itself, is only applying
the model. The logic to do
something interesting with
that model and data still needs
to be programmed and housed
somewhere onsite.
One interesting application
has cropped up recently.
Since the advent and massive
adoption of IP cameras
throughout intelligent
buildings, a few smart data
scientists saw they could
do even more. There is now
software that exists that can
in real time identify whether
someone is holding a real
pistol or a water gun automatically without the need for
humans to inspect it. These types of systems are starting
to spring up in high schools to thwart any potential
active shooters. This is only possible due to recent
machine learning advances that are only possible
because of the sheer amount of data that these
cameras are collecting. Smart building indeed! The
linear journey from big data to machine learning to
the edge has surprised many in ICT, IT, and operational
technology (OT), even though in hindsight it is
only logical.
Another big area for applying machine learning
in building management systems is with the notably
complex HVAC systems. Good engineers can tune
these systems by hand using the data they have
available, but there is no way they can continuously
tune to adapt to real-time data that streams in from
thousands of given inputs.
Calculating the thermal load in a room can be
done using an equation with quite a few parameters.
Machine learning in this case is finding the optimal
values for each of these variables in an automated
fashion. The decision from those values can be tested
and learned over time thanks to the copious amount
MACHINE LEARNING MAGIC
The entire field of inference includes a wide range of
activity that ten years ago would have been viewed
as magical. Automated image classification looking for
defects in wind turbines, acoustic detection that can
“‘hear” manufacturing lines operate in ways that
would make dogs blush, not to mention maritime’s
need for detecting, tracking and classifying marine
vessels. Currently, there are systems smart enough
to understand what type of vessel is out there in the
world’s waters simply by ”listening.”
Predictive maintenance is already in place in
industries, such as oil and gas, to predict workover,
rod change, and cleaning operations leading to
drastically reduced downtime and preventing
catastrophic repair costs. Of course, seismic
interpretation can be enhanced here as well for salt
classification. Parted rods or leaking tubing can
leave pumps not pumping for days or weeks.
A new technology is what some people are calling
”super sensors.” Super sensor technology is when
regular sensors cross-breed with machine learning
to detect complex events. As the more granular these
super sensors get and the larger the supply chain