MACHINING
PUTTING AI IN MAINTENANCE
NOVOTEK
MACHINE LEARNING IS STREAMLINING PREDICTIVE
MAINTENANCE
Over the past five years, the industrial sector
has begun to see the value in digitalisation
and has invested more in adopting it. With
this has come a cultural shift from reactive
equipment maintenance to proactive
maintenance that pre-empts problems.
Here, Sean Robinson, service leader at
industrial analytics platform supplier
Novotek UK & Ireland, explains how plant
managers can make proactive maintenance
even more effective with machine learning.
In 2006, UK mathematician Clive Humby
claimed that “data is the new oil”. Whether
you’re a food processing company or
an automotive manufacturer, data from
production processes is the cornerstone of
better efficiency, effectiveness and overall
performance. Plant managers that are
familiar with the industrial internet of things
(IIoT) will know that one of the concept’s
biggest selling points has been the insight
it can provide into equipment performance
and process effectiveness, which in turn
creates benefits for the company’s bottom-
line.
This has changed the culture of maintenance
in plants that have started adopting IIoT
technology. Rather than responding
to a breakage or conducting planned
maintenance based on expected equipment
lifespan, engineers can make informed
decisions about when to maintain systems
based on the equipment’s condition.
Minimising unplanned downtime has
obvious benefits, but it’s the reduction in
scheduled downtime that adds significant
value in terms of increased overall
throughput for no new capital outlay.
However, achieving this is challenging due
to the volume of data and subsequent
analysis that is required to confidently
change maintenance schedules.
This is where an opportunity arises for
machine learning in industrial maintenance.
With machine learning, algorithms can be
trained to identify correlating factors in data
to not only flag up a problem but also the
root cause of it. It sounds straightforward in
principle, but the number of potential things
to consider can be too high for a human to
work through effectively.
Within a single machine, there can be
dozens of sensors or other health signals.
90
PECM Issue 33
To get a clear picture of all the things
that affect reliability, that data should be
evaluated alongside things like maintenance
records and a history of what the machine
was running.
Even ambient conditions and crew data can
give clues as to what issues can crop up.
The only effective way to navigate the
abundance of variables is with an IoT
platform with machine learning, such
as GE Digital’s Predix platform and Asset
Performance Management (APM) suite.
Connecting an IoT-enabled machine to the
platform allows Predix’s machine learning
algorithms to analyse it with the APM’s
combination of standard measures and
advanced analytics.
This allows maintenance staff to not only
spot when a machine needs maintenance,
but also why.
For example, a semiconductor manufacturer
might find that it rejects ten per cent of its
output due to faults in the manufacturing
process. Although all the machines may be
IoT-connected, there is too much data for an
engineer to reasonably analyse.
With Predix’s machine learning algorithms,
the APM could, for example, identify that a
machine has elevated vibration levels, which
is damaging the semiconductors.
The algorithms can then assess this against
historic data to spot patterns in how often
this occurs, identify the performance
signs that precede it and — if integrated
into a management system — send alerts
to engineers as the machine requires
maintenance. This makes it possible for t he
machine to receive maintenance only when
its conditions indicate it should, changing
from preventative to condition-based
predictive maintenance.
In effect, machine learning allows
maintenance data analysis to become a
more automated process. In fact, there are
certain industrial applications where the
algorithms could be permitted to directly
reconfigure a machine with the right
settings. And as machine algorithms learn,
this will become an increasingly viable way
of improving efficiency.
Whether you believe data is the new oil
or not, it’s indisputable that it’s a valuable
resource that fuels overall operational
improvement for plant managers and
maintenance engineers. The key to
achieving this is to use industrial analytics
intelligently and effectively to strike oil in
industrial maintenance.
www.novotek.com/en/