CO2_2016 - page 33

Chimica Oggi - Chemistry Today
- vol. 34(2) March/April 2016
milling time and does not provide enough material to meet the
demands of LCD manufacturers.
If the milling process cannot be operated to the required
speed of 5 hours or less to mill down to 200 nanometres, it will be
necessary to incur the capital cost of purchasing a second milling
machine, as well as tolerating the ongoing cost of an energy
inefficient process. This will impact profitability and the availability
of capital to invest in other projects.
Milling is carried out in a horizontal bead mill. This is a chamber
filled with beads, through which the dispersion of pigment is
passed. The beads in the chamber are agitated at high speeds
to grind down the pigment particles. The potential factors
affecting the efficiency of this process are numerous and include
those relating to the mill itself, the method of operation, the
formulation of the dispersion and the milling media used.(7)
Therefore a solution requires first understanding which, out of a
large number of candidates, are the important factors.
Analysis of data from the prior 17 production runs using a
bootstrap forest in figure 9 identified a Pareto ranking of the top
factors to investigate with DOE.
The potential factors identified in Figure 9 were
reviewed from a scientific perspective to make
a final decision about the factors to progress into
the DOE. This resulted in the top nine variables
being selected excluding Temp(Out) and P1.
A Definitive Screening Design(4) in seven
factors was designed which resulted in 17 DOE
combinations. This was deemed the most efficient
way of progressing as it allows screening for the
critical factors, and providing three or fewer
factors are important then interactive and non-
At the start of the DOE learning process you may have existing
data, in which case effective statistical modeling of that data
may aid the design of the next DOE. In particular, analysis of prior
data may help decide which factors to include in your DOE and
the range over which to vary them.
Existing data may be messy which makes it difficult to correctly
extract information on the factors and factor ranges. Messy data
issues may include the factors being related, e.g. an increase in
one factor results in increase or decrease of another factor, some
of the data cells may be recorded incorrectly, other cells may be
empty or missing. In these situations data mining methods, such
as the bootstrap forest (or Random Forest
)(5) method available
in software packages such as JMP Pro,(6) may help extract the
potentially important factors from a set of correlated factors
using approaches that are robust to data errors and missing cells
to provide insight about the potential factors and factor ranges.
Integrating such methods with modern DOE approaches, we
gain reductions in total learning time, effort and cost.
A specialty chemicals supplier of pigments to liquid crystal
display manufacturers is experiencing problems in manufacturing
enough pigment to the required specifications to meet customer
demand. To achieve sharp displays, the pigment particles
must be milled down to less than 200 nanometres and the
time taken to do this is extremely variable. The milling stage is
energy-intensive and a bottleneck, the long mill time is incurring
excessive energy cost and is impacting throughput. To meet
product demand requires a faster process or additional milling
equipment to run in parallel with existing equipment.
Figure 8 shows an upward trend in the time to mill to less than 200
nanometres for recent production batches of pigment. To avoid
the capital cost of adding additional milling equipment to meet
product demand, we need to get mill time to below 5 hours,
which appears to be a challenge as none of the prior batches
has achieved this goal.
OFAT was used to develop the process initially, but because
the process is taking too long, OFAT has resulted in the need to
repeat earlier development work to try and learn how to speed
up the process. In addition to the cost of repeating prior cycles of
learning, the current process incurs high energy costs due to long
Figure 8.
time to mill
of first 17 batches.
Figure 9.
Input Variable
1...,23,24,25,26,27,28,29,30,31,32 34,35,36,37,38,39,40,41,42,43,...68
Powered by FlippingBook