How Democratized Artificial Intelligence Can Move Manufacturing to a New Evolution Pace
specific machine on high and low
temperature, vibration and humidity
indicators (by the way, you can now
anticipate the future of them using AI
predictive tools).
parallel, different algorithms to provide
different probabilities of selecting of the
best AI algorithm which fit to the data and
helping to eliminate non-relevant data from
the dataset; mostly in case of supervised AI
models.
But it is not enough if you want to detect
potential issues: You may have to detect a
repetitive temperature and humidity
fluctuation hidden in the normal behavior of
the equipment which could have an impact
on your production efficiency and quality.
The aim is to automatically detect this
unknown repetitive pattern to understand
why and correlate this abnormal behavior
with the potential side effects, using
techniques like Grubb’s test 17 , SAX (Symbolic
Aggregate approximation) 18 or LOF (Local
Outlier Factor) 19 . The most important is that
business applications embedding AI are
helping based on the business and
application domain and by computing, in
The second case could be to detect, for
instance, one small fluctuation: While
usually the temperature fluctuation is
medium, a small fluctuation should generate
at least a warning to investigate and take a
corrective action. The same as predictive AI,
the new real-time capability provided by
new
Information
Technology
(IT)
technologies (Graphics Processing Unit
(GPU), In-Memory, Cloud and High-
performance communication) is making this
feature available to make fast decisions in
day-to-day activities. Of course, this
technology can be used for many topics like
operator mistake detection, raw material or
Figure 3: Example of anomaly detection application for industrial equipment
17
https://en.wikipedia.org/wiki/Grubbs%27_test_for_outliers
18 https://jmotif.github.io/sax-vsm_site/morea/algorithm/SAX.html
19
; http://www.cs.ucr.edu/~eamonn/SAX.htm
http://scikit-learn.org/stable/auto_examples/neighbors/plot_lof.html
IIC Journal of Innovation
- 11 -