Page 48 - Practical Control Engineering a Guide for Engineers, Managers, and Practitioners

Introduction to Developing Control Algorithms 23

located at various points in the block diagram that have similar short-
term trends due to these disturbances and malfunctions. A good proj-
ect manager can have both approaches active and complementary.

Time Domain Analysis
Now that a module has been identified and the specifications gath-
ered, it is time to "look" at the process in the simplest most logical
way-in the time domain. This means collecting data on selected pro-
cess variables local to the module and studying how they behave
alone and when compared to each other. Before starting to collect the
data the team should agree on the key process variables to collect and
on what frequency to sample them. This may require installing some
new sensors and even installing some data-acquisition equipment.
Decades ago, the only source of data was the chart recorder.
Nowadays, most processes have computer-based data-acquisition
systems, many of which not only collect and store the data but can
also plot it online. These systems can also plot several process vari-
ables on the same graph. The opportunities to look at the process
dynamics in creative ways are nearly endless. Use your imagination.
Gaining insight and solving problems are the primary goals of
the activities associated with each of the four comers of the diamond
in Fig. 2-1. The time domain plots will likely reveal problems that
should be solved by the team (as soon as possible) thereby reducing
variation in the local process variables connected with the module.
Reducing variance locally is the immediate challenge. Do not worry
about the impact of these activities on the end-of-line product vari-
ance. That will come later.
Frequency Domain Analysis
Once the time domain analysis/problem revelation/problem solving
has begun, it often makes sense to look at the process module in some
other domain. The road-map diagram shows a second corner labeled
"Frequency domain analysis." Here, without going into too much
technical detail, one uses Fast Fourier Transform software to develop
line spectra or power spectra for selected variables. Essentially, long
strings of time domain process data are transformed to the frequency
domain where sometimes one can discover heretofore unknown peri-
odic components lurking in noisy data. Few computer-based data-
acquisition/ process-monitoring systems have the frequency domain
analysis software built in, so the engineer will have to find a way to
extract the desired process variables and transfer them to another
computer, probably off-line, for this type of analysis.
Figure 2-3 shows a long string of time domain data for a process
variable. The variable was sampled at a rate of 1.0 Hz (or every second).
In the time domain it simply looks noisy and seems to drift tightly
around zero (perhaps after the average has been subtracted). When
transformed into the frequency domain, Fig. 2-4 results. Here the