Chemical Engineering June 2018 - 83

FIGURE 4. A sophisticated analytical platform (compared to just the use of
a simple Excel spreadsheet) can help users capture the CPPs easily, pulling
them out of the historian automatically
FIGURE 5. The data analytics platform can easily create superimposed graphs
using automatic ranging and scaling capabilities. This allows process engineers
to quickly recognize which curves tend to form a tight group, and the
platform can aggregate all of these curves into an ideal profile for that CPP.
This becomes the reference profile for that CPP for that product manufactured
using that equipment
A more methodical approach
To solve the puzzle, an enterprising
process engineer may gather
up data from the reactor's control
system and data historian (Figure
1). Loading it into Microsoft Excel
or similar software is a tedious but
manageable task to extract data
and create graphs (Figures 2 and 3),
and to compare specific attributes
applications make analysis easier
by extending number-crunching capabilities
beyond a normal spreadsheet,
combined with mechanisms
to aggregate data.
Ongoing technology advances
Ongoing advances in analytics software
have greatly simplified the
infrastructure required to impleProcess
operators can advance theories of what might
be happening, and use data-analytics tools to see if the
operating experience supports a potential cause-andeffect
relationship. Today's tools allow such theories to
be tested with relative ease
tied to CPPs across multiple identical
unit procedures.
Batch by batch, it is possible to
examine the solution temperature,
the agitator speed or other factors
that might be suspected of causing
a problem. This approach sometimes
works, and it might even generate
some possible answers, but
it will probably be necessary to dig
deeper. The practical limits of manual
analysis with Excel for complex
processes will soon be encountered.
The next level is a more sophisticated
data-analytics approach. This
may be a function built into a data
historian, or more typically, a thirdparty
application designed to work
with data from the historian. These
74
ment data-analytic applications.
Cumbersome techniques, such as
spreadsheets, data cubes and data
warehouses, are no longer needed.
Modern data-analytic platforms can
run on typical
office
computers,
communicating directly with the historian
and extracting data as needed
for any computations.
Returning to the example discussed
above, process engineers
watch six CPPs connected to unit
Procedure A. Using data recorded in
the historian from the 60+ unit procedures
(Figure 4) that generated
good batches with acceptable specifications
on all CQAs, it is a simple
matter to graph the six variables for
all the unit procedures. Curves for
each CPP from all 60 unit procedures
can be superimposed on each
other using identical scales.
Using this approach, it is easy to
see whether the curves tend to form
a tight group, or if they spread out,
showing different values at various
times for a given CPP (Figure 5).
The data-analytics platform can aggregate
all these curves into an ideal
profile for that CPP. This then becomes
the reference profile for that
CPP, for that product manufactured,
using that equipment. This is necessarily
specific, given the importance
of only comparing " like data. "
Using the reference profile, it is
possible to generate upper and
lower tolerance limits - or boundaries
- based on an appropriate
value, such as two or three standard
deviations (Figure 6). Exactly where
these boundaries need to be set will
ultimately be determined by the process
and how variability of this CPP
affects the final product, but using
this approach is a good place to
start. There is no advantage in making
every tolerance profile equally
tight. Allowing wide tolerances for
variables that still support CQA requirements
will ultimately create a
process that is easier to control.
This same procedure can be repeated
for every CPP to create a reference
profile and a set of boundaries
for each of those variables. Data
for a CPP can even be aggregated
into a single value, which can become
a useful metric for comparing
many batches. Histograms of these
metrics provide a quick view of process
variability.
When the process is thus charCHEMICAL
ENGINEERING WWW.CHEMENGONLINE.COM
JUNE 2018
http://WWW.CHEMENGONLINE.COM
Chemical Engineering June 2018

Table of Contents for the Digital Edition of Chemical Engineering June 2018
