Glass Bulb Production
Object of Investigation
The project was performed for GE in 1991.
The object of investigation was to review a production
line containing a furnace, transportation line, and the bulb formation
unit (producing special bulbs for nuclear reactors). Two-hundred (200)
different parameters were measured at 3 minute intervals for quality
control. These measurements consisted of physical and operational
variables The parameter used to define bulb quality was the geometric
The main goal was to determine a set of variables that
essentially predicted the bulb size.
The data set was considerably noised, and the preliminary
(conventional) statistical analysis revealed no noticeable relationships
in the database. The opinion of plant engineers (based on their
intuition) indicated a satisfactory prediction was possible.
The KET analysis did confirmed that a prediction was
possible. The results discovered that only 15 of 200 variables were
essential to predict the problem. The constructed predictor demonstrated
a 93% accuracy. Additionally, there were two major database shortcomings
identified: a lower frequency of quality control measurements (that
parameters every 0.5 minutes were more desirable) and presence of (human
factors) artifacts in the database that interrupted normal
The economy obtained from reducing the frequency of
measurements was an immediate gain. The opportunity to use a quality
predictor and apply it to quality control, promised potential
improvement in the manufacturing process.
it was Done
An Information Analysis Module from the KET Tool
Kit was used to measure the dependency between variables, identify and
list the essential ones. This type of analysis has great advantages
comparatively with the classical correlation theory, because it does not
require any assumptions about the underlying models (as the correlation
analysis inherently does).
Subsequently, Topologic Analysis was used to assess the
data informativeness. It discovered shortcomings in the database and
assess the potential contribution of different parameters before the
actual predictors were constructed. Finally, the predictors were
constructed with the help of the Fast Analytical Descriptor from
KET’s “Model-Free,” Tool Kit, instead of using a
predetermined one and simply fitting the coefficients).
The further applications of KET may include the creation
of a control/monitoring system with diagnostic, decision making and data
collecting modules. The system was designed to adaptive to technological