86 SMT007 MAGAZINE I AUGUST 2020
the other important measure of productivity is
on the gross retest count. This is simply a total
of how many times any board is retested.
For every retest, an operator must perform
physical manual tasks that would take a few
minutes. A percentage (typically between
10–30%) of these total retests would require an
engineer to investigate further, which will take
another 5–10 minutes of manual task. Every
factory would assign a fixed cost per minute for
these wastages. As such, it is tangible to calcu-
late the savings in terms of cost per month due
to retest. Figure 12 shows the actual monthly
total retest count tracking. It is evident and
reassuring that since applied ML of anomaly
detection was implemented, the gross monthly
total retest count has improved by as much as
19%. This directly translates to more than 300
hours of additional productivity per month.
Conclusions
Advanced analytics techniques, such as
anomaly detection, combined with a big
data infrastructure for real-time data acquisi-
tion and ingestion—along with a clear vision
and digital transformation strategy from man-
agement—can be significantly beneficial to
most high-value factories. These techniques
require complicated measurements, and a vast
amount of them, to define the product qual-
ity and quantity. The key to a profitable ROI is
successfully mapping the predictive outcomes
to potential tangible root causes for the pro-
duction team to take decisive action before the
problem escalates. SMT007
References
1. S.J. Russel & P. Norvig, Artificial Intelligence: A Mod-
ern Approach, Prentice Hall, December 1994.
2. M. Walker, "Hype Cycle for Emerging Technologies,
2018," Gartner Research, August 6, 2018.
3. Scikit Learn, "Metrics: F1 Score."
4. Scikit Learn, "Metrics: Matthews Correlation Coeffi-
cient (MCC)."
5. Scikit Learn, "Metrics: Average Precision Score."
6. Scikit Learn, "Metrics: Recall Score."
7. Scikit Learn, "Metrics: Outlier Detection With Local
Outlier Factor (LOF)."
8. DeepLearning4J, "Basic Autoencoder: Anomaly
Detection Using Reconstruction Error."
Derek Ong, BscEE, is a
manufacturing technology and
software solutions planning
manager for Keysight Technologies.
Figure 12: Tracking gross retest count.
