Maintenance Technology June 2017 - 16
Predictive analytics and machine-learning algorithms are sharper than human eyes.
software then examines the data
to determine root causes and
early-warning indicators from past
downtime issues. Finally, the analytics
software develops and deploys "agents"
that monitor data traffic either locally
or in the cloud.
Analytics software uses two types of
agents. The first type is failure agents,
which watch for patterns that are
known to predict a future failure. If
such patterns are detected, the agents
alert plant personnel and deliver a
The second type is anomaly agents,
which watch normal operating
patterns and look for changes, such as
operating or environmental-condition
changes. These agents also alert
personnel of any detected changes so
they can investigate and take corrective
action if necessary.
Your crystal ball
While technicians may not be able to actually 'see' into the future, smart technologies
and advanced analytics can help them predict it.
on the incident, which involved
reviewing 16 months of data, found
that the bearing cooling system had
not been operating correctly for six
Had this data been used as part of
a PdM strategy, the company likely
would have been able to identify the
bearing degradation and its root cause
before the failure actually happened.
What's more, the company would
have been able to identify detailed
preventive-maintenance steps for the
Predictive maintenance also can be
valuable in operations that experience
high maintenance costs.
Often, companies can invest a lot
of time and resources in maintenance
but lack data to know whether their
strategy is effective and addressing
their actual needs. Predictive maintenance can help uncover unnecessary
maintenance, which could save
millions of dollars every year in some
industries. This was another discovery
in the compressor case. The company
was performing certain maintenance
activities that were unnecessary and
could have been eliminated.
How it works
Predictive maintenance doesn't require
an extensive overhaul of your infrastructure. Rather, it can be deployed on
your existing integrated-control and
The process begins with discussions to identify what data you want
to collect, what potential failures or
other issues you want to predict, and
what issues have arisen in the past.
From there, the relevant historical data
is collected from sensors, industrial
assets, and fault logs.
Predictive technology has been
around for decades. It's used to detect
credit-card fraud, fine-tune marketing
programs, and even help us search the
Internet. Its role in the industrial world
takes the form of a rigorous documentation of events and failures that
can help us see and address machine
or equipment issues in their earliest
Many manufacturers already see the
value of historical failure reports as a
tool to help prevent failures and downtime in the future. By using this data,
which already exists in your assets, you
too can reduce downtime surprises,
cut down unnecessary maintenance,
and potentially reduce risks in your
Information for this article was provided
by Doug Weber, engineering manager,
and Phil Bush, remote monitoring
and analytics product manager,
Rockwell Automation, Milwaukee.
For more information, visit