Outcomes, Insights and Best Practices from IIC Testbeds: Smart Factory Machine Learning for Predictive Maintenance Testbed
where the end user can support their
decisions to change it during the next
maintenance stop. Usually, the result or
output of machine learning algorithms is
quite complex and requires a lot of
experience to properly interpret the
results—arguably no other field of
experimentation requires the feat of
knowledge that the end user needs for
machine learning algorithms. It is also
important to consider that the actionable
insight given to a machine operator would
not be the same as the insight given to the
line manager of a production facility.
pull sensor readings from different places,
which is a form of sensor fusion. Taking
advantage of this element’s Field
Programmable Gate Arrays (FPGAs) helps to
accelerate the machine learning algorithms
and is another technology the testbed aims
to develop further. The hardware being used
acts as a platform in which the new
predictive maintenance technology can be
deployed. Regarding software, the testbed
works with different protocol technologies
related to industrial parts. It also
incorporates Industrial Internet of Things
(IIoT) technologies which help transport the
data. OPC Unified Architecture (UA) is one
example of this IIoT utilization. The Data-
Distribution Service for Real-Time Systems
(DDS) standard, as implemented in DDS-
Secure from RTI , is another example
included in the testing.
The Smart Factory Machine Learning
Testbed is currently working on various use
cases, one of which involves the spindle
head of a Computer Numeric Control (CNC)
machine tool used to manufacture
crankshafts for the automotive industry. The
spindle head is the most difficult part for
which to predict the failure of internal
elements. Other use cases of the testbed are
related to failure points such as ball bearings
and ball screws, where behaviors and
patterns in energy consumption are used to
support the decision making. The energy
data can be fused with other types of data
coming from the machine to solve specific
problems. Different use cases will be
addressed in the near future as the next
phase of the testbed addresses problems
with surface heat treatment. The testbed
will need to detect and analyze particular
failures of a critical element in a laser heat
treatment process.
The Smart Factory Machine Learning
Testbed is deployed over highly sensorized
machines which are nearly autonomous.
There is a sufficient amount of data coming
from different sensors already in place. For
state-of-the-art industrial machines, the
acquisition of the energy consumption is
done at a relatively low frequency, around
three kilohertz. The testbed is deploying a
new sensor to measure energy. In one
platform, the sensor might measure eight
kilohertz to 32 kilohertz. In this unique case,
the testbed must detect the small deviations
or sparks that occur during processing and
analyze the output pattern to understand
whether something is changing to affect the
energy consumption of the production
element.
Many unique technologies have been
utilized in the testbed’s use cases. In terms
of hardware, the testbed is working with
Xilinx’s UltraScale™ MPSoC Architecture to
The testbed is deployed in three locations.
The first is in the Aingura IIoT labs in Spain,
- 116 -
November 2019