Design of Experiments and Statistical Process Control - Fundamentals of Industrial Chemistry: Pharmaceuticals, Polymers, and Business - page 232

design, and perform a fraction of the experiments. A half-factorial design of
a ive - factor design would involve 16 experiments, half of the 32 needed
for a full-factorial design. With a half-factorial design, the main effects and
first-order interactions can still be determined. If there is any ambiguity, fur-
ther experiments can be subsequently performed. In 2011, as part of their
Quality by Design initiative, the FDA encouraged design of experiment tech-
niques in new drug applications with the goal of improving the quality of
drugs and ensuring supply-chain safety[1].
One variation is called Evolutionary Operation of Processes (EVOP). This
technique involves studying small changes to a process during normal oper-
ation, to find the optimum operating conditions. The advantage is that it can
use data already being generated by an existing process and has minimal dis-
ruption on manufacturing.
13.3 STATISTICAL PROCESS CONTROL
Processes must be controlled for quality purposes. Most major chemical pro-
cesses have automatic computerized data logging and many parameters are
automatically recorded and available for study. These can include set points,
process measurements such as flow rate, temperature, pressure, stirring speed,
and so forth or product parameters such as composition, density, clarity, and
so forth, especially when the product is measured on-line.
Let's consider a simple process to make a formulation containing 50% by
weight of component A. Component A and the other ingredients are contin-
uously fed from different belts into a mixer. The blend continuously exits the
mixer and is measured for the weight percent component A in the blend. For
the sake of this example, let's assume that the measurement is made once a
shift or once every eight hours. The process operators have the option to adjust
the feed belt speed and therefore the amount of each component fed to the
mixer. If the blend goes out of quality, they are expected to make these adjust-
ments so that the product is suitable for sale. The process has been running
fine. Imagine that it runs for almost another week with no process adjustments
and with the following results (Figure 13.8):
Had you been the operator during this run, would you make any adjust-
ments? After the first measurement of 48%, which is lower than the desired
50%, it might be tempting to adjust the belt speeds to increase the feed of
component A. However, with no adjustment, the second result was 51.3%.
With an adjustment it likely would have been much higher and therefore fur-
ther from the target of 50%. How did the result go from 48% to 51.3%? Is it
significant? Is it reflective of a need for a process change such as a change in
the feed belt speed or is it normal process variation. Or is it variability in the
test measurement? Or a combination?