Control is a hidden, enabling
technology that is present in almost every engineered system today. Despite
this fact, control system design is still mysterious and often falls in the
domain of a specialist. Today, every engineer must know how to design, implement
and integrate a control system into a design from the start of the design
process. An engineer needs to understand how to balance performance, low cost,
robustness and efficiency to effectively accomplish these goals.
Evaluating a design concept is
best done through modeling, not by building and testing, as modeling provides
true insight on which to base design decisions. There is a hierarchy of models
possible of varying complexity and fidelity, but a simple design model which
captures essential attributes is the most useful, i.e., dominant dynamics. An
integrated control system can enhance a design through stabilization, command
following, disturbance and noise rejection, and robustness. All of this can be
accomplished through a combined approach, rather than trying to accomplish all
with a single feedback controller, as is too often the case.
To best understand this
combined approach, I had extensive discussions with Dr. Rob Miklosovic, a
leading mechatronics innovator at Rockwell Automation in Cleveland, OH. The
diagram illustrates just such an approach.
Click here for a larger version
The design model is typically
used for both feedback and feedforward controller design. However, in practice,
the physical system will deviate from that design model. A disturbance observer
regards any difference between the physical system and the design model as an
equivalent disturbance applied to the model. It estimates the disturbance and uses
it as a cancellation signal. So in addition to enhancing disturbance rejection,
the disturbance observer makes the physical system behave like the design model
over a certain frequency range, thereby simplifying the design of the feedback
and feedforward controllers. Since the design model inverse is not realizable,
a unity-gain, low-pass filter, specifying the observer bandwidth, is added.
Next, the feedback
controller is designed solely to force dynamic consistency by mitigating the
effects of model uncertainty and disturbances, usually with high gain and
integral control. A common mistake is made in designing the feedback controller
for desired output with no regard for robustness, only to find poor performance
when applied to the physical system. However, once consistency is enforced, the
desired output can be augmented with a feedforward controller, typically the
dynamic model inverse, to recover the dynamic delay of the closed-loop system
with no effect on stability or properties of the closed-loop system.
The combination of a
disturbance observer with both feedback and feedforward controllers is not new
and many researchers have demonstrated its effectiveness. What needs to be done
now is to bridge that theory/practice gap and put this technique into the hands
of the mechatronics
engineers responsible for creating the innovative systems
we all need.