Page 35 - Practical Control Engineering a Guide for Engineers, Managers, and Practitioners

10 Chapter One

strong emphasis on the feedforward component. At the risk of over-
simplification, driving a car depends heavily on the driver looking
ahead, noting changes in the road and traffic, anticipating distur-
bances, and making adjustments in the steering wheel, gas pedal, and
brake pedal. The actions taken by the driver are the result of many
months and sometimes years of training and constitute a human
feedforward algorithm. There are human feedback components to
these feedforward adjustments but they are mostly corrections for
inaccuracies (hopefully small) in the training and experience that
constitutes the human feedforward algorithm.
If the automobile were to be driven exclusively by feedback con-
trol, the right-hand side of Fig. 1-13 shows that the driver could not
look out through the windshield. Instead, the driver must make
adjustments based only on information gathered by looking at the
road through a hole in the floorboard. This kind of restriction would
force the driver to maintain a slow speed. Here the driver is carrying
out feedback control and is able to react only to current disturbances
and has no information on upcoming disturbances.
Consider the case of driving down the center of the road by follow-
ing the white line as seen through the hole in the floorboard in the face of
strong gusting crosswinds. Since this is a hypothetical question, put
aside the obvious fact that this activity would be illegal and dangerous.
One can surmise that a strategy of reacting aggressively to short-term ran-
dom bursts of wind to keep the white line precisely in the center of the
floorboard opening would probably put the car off the road. Instead,
because the disturbances are not constant but unpredictable, the driver's
best strategy might be to conservatively adjust the steering wheel to keep
the white line, on the average, "near" the center of the floorboard opening
and tolerate a reasonable amount of variation. Therefore, rather than
react to sl10rt-temt variations, the driver would have to be content with
addressing /ong-temt drifts away from the white line. I recently drove
from New York to Colorado and back. I found myself reacting to sus-
tained bursts of crosswind in a feedback mode. Therefore, the arguments
of this section suggest that the sustained bursts of crosswind might not
be classified as unautocorrelated.
Based on this rather extreme example, we can perhaps conclude that
using feedback control on a noisy industrial process will probably not
produce perfect zero-error control. Since feedforward control is rarely
available for industrial processes, if one really wants to decrease the
impact of short-term nonpersistent disturbances, he must actually "fix"
the process, that is, minimize the disturbances affecting the process.

1-7 An Example of Controlling a Noisy Industrial Process
To illustrate the impact of feedback control on noisy processes, con-
sider a molten glass delivery forehearth shown in Fig. 1-14. Since the
reader may not have a glass-manufacturing background, a little