MAINTENANCE MAY/JUNE 2020
Manufacturers can embrace a deep learning approach to help
avoid unplanned maintenance
“Machines today work
in symbiotic harmony
with each other”
Rob Dalgety, industry specialist, Peltarion
www.manufacturingmanagement.co.uk
22
BY ROB DALGETY, INDUSTRY SPECIALIST, PELTARION
The manufacturing industry has been
going through an enormous amount of
change, thanks in large part to digital
transformation. Today’s smart factory
combines the physical with the digital,
with entire assembly lines operated
by a new generation of robots, IoT sensors and
software. Sensors collect data on every aspect
of the factory, helping to ensure processes are
scalable, fast and repeatable. The next wave of
innovation will build on these blocks and propel
manufacturing even further, integrating these
mechanisms with Artifi cial Intelligence (AI).
Take a proactive response to avoid
unplanned maintenance
Machines today work in symbiotic harmony with
one another. If one machine has a maintenance
issue, every hour, minute and second counts. The
complex nature of today’s business means that if
a machine needs to be taken out of circulation,
it must be carefully managed so that it doesn’t
create a chain of events that disrupts factoryline
production – or even cascades down the
broader supply chain. Being able to manage
any maintenance work in an ordered way is
far less disruptive or costly. In fact, unplanned
maintenance can cost between three and nine
times more than planned maintenance, due
to overtime, call-outs and parts being needed
quickly – not to mention
revenue leakage from stopped
production, fi nes due to broken
SLAs and reputational damage.
One goal of any
company is to reduce the
number of incidents of
unplanned maintenance
by carefully controlling
planned maintenance. Many
manufacturers will replace
and service machinery using a
set timetable. This can be very
eff ective in reducing unplanned
maintenance issues, but it will
not account for anomalies.
Many manufacturers (and other
businesses with machines in
their operations) have no idea
that there is a problem with
a machine until it breaks, by
which time it is already too late.
Planned maintenance
Deep learning is well suited
to the task of predictive and
proactive maintenance, because
of its innate ability to identify
patterns in scenarios involving
large, complex datasets
containing multiple types of
data.
Most modern factories
are already outfi tted with
sensors, off ering a wealth of
information on every aspect
of the factory. Deep learning
can add power to predictive
maintenance data by enabling
data scientists to add multiple
Manufacturers
are urged
to think
smarter about
maintenance
types of data to models
(including images, audio and
video) on top of existing sensor
data. For example, audio data
is a powerful source of data for
predictive maintenance models.
Sensors can pick up sound and
vibration, which is then used in
the deep learning models. This
data can be the most critical in
determining machine life and
degradation cycles.
Also, unlike other machine
learning methods, the more
data you can provide, the
more deep learning models
improve and the more accurate
the model becomes. This
enhanced dataset then powers
a comprehensive predictive
model, helping manufacturers
to, for example:
1Reduce operating costs,
and increase reliability and
productivity on factory fl oors;
2Monitor complex power
grids that contain enormous
numbers of diverse assets
such as transformers, cables,
turbines or storage units;
GOING
DEEP
50%
of manufacturers plan
to use machine learning
for maintenance
/www.manufacturingmanagement.co.uk