Page 44 - Practical Control Engineering a Guide for Engineers, Managers, and Practitioners

Introduction to Developing Control Algorithms 19

Far more frequently, a priori mathematical modeling simply is
not up to the task. Most industrial processes are just too complex and
contain too many unknown idiosyncrasies to yield to mathematical
modeling. I have more to say on this problem in Sec. 2-4.

2-1-3 A Common Sense, Pedestrian Approach
If the process exists and is accessible, the control engineer adds extensive
instrumentation, studies the process using the methods presented next,
and, if necessary, develops an algorithm from the process observations.
When the process is not accessible, one makes a heavily instru-
mented prototype of the process and develops a control algorithm
around the empirical findings from the prototype.
Alternatively, if it is a new process, yet-to-be-constructed, and a
prototype is not practical, the engineer negotiates for added monitor-
ing instrumentation. In addition and, even more difficult, he negotiates
for up-front access to the process during which planned disturbances
will be carried out so that one can find out how the process actually
works dynamically. During this up-front time, many unexpected prob-
lems can be discovered and solved. The control algorithm vehicle, usu-
ally digitally based, is designed with extensive input/output "hooks"
for diagnosis. Finally, the control algorithm is designed around these
findings. I have frequently made mathematical models based on the
empirical evidence gathered during these up-front trials.
This approach is significantly more expensive in the short-term
and often violently unpopular with project managers. I have consis-
tently found it to be a bargain in the long-term. There is some style
required here; the engineer must convince the management that the
extra instrumentation and up-front learning time is required. Junior
control engineers usually are not aware of this approach-mostly
because they have not yet experienced the disasters associated with
SMILH and a priori methods. But, even if they are aware they usually
cannot convince a seasoned project manager about the benefits of
taking a pedestrian approach simply because they haven't a track
record of success in this area.
If the process for which the control algorithm is to be developed
already exists then this empirical approach is really the only valid
choice IMHO. Since this case is so prevalent and special it will com-
mand a whole next section.

2-2 Dealing with the Existing Process
Consider the following scenario. A section supervisor in a manufac-
turing plant is not satisfied with the performance of the process for
which he is responsible. The end-of-line product variance is too high.
Thinking that the solution is more or better process control, he calls in
the control engineer.