Introduction to Feedforward Control Video Lecture - Electrical Engineering (EE)

In this screencast we will talk about feedforward
control and its role in developing appropriate
control schemes. So before we can really talk
about where feedforward control can be used
we have to first focus on where feedback control
fails. Because a large majority of control
processes are of a pure feedback only control
scheme, so therefore in order to deviate from
those schemes there have to be reasons why,
and there are a number of reasons why, but
here we'll focus on three of them. The first
is that perfect control cannot occur. A disturbance
happens, but the controlled variable does
not feel it, so in other words a disturbance
to the variable will happen, but the control
process will already handle it. Because feedback
control does not work until there is a deviation
to the controlled variable, it is impossible
for a disturbance to have no effect on this
variable, which connects number two, because
you are only measuring the controlled variable
at the end of the process, there is no way
to predict what's going to happen to this
variable. An additional problem is that processes
are slow or have large dead time can lead
to control, which can take a while. Why? Because
if a system has a very large dead time, for
example, a disturbance will occur, the process
won't feel it for a while, but once it feels
it, it's going to take a while to actually
act on it because of the slowness of the process.
So feedforward control is a way to handle
all three of these drawbacks. So to try to
represent this we are going to look at a somewhat
rudimentary example of a blending tank, and
in our blending tank we are trying to control
the output concentration, and we have two
streams, so therefore an example of a feedback
only scheme would be one where we would try
to directly control the composition coming
out by the flow rate of the first one. However,
one example here would be let's say that the
flow rate of the first scheme is known to
fluctuate a bit, so therefore if that is a
potential disturbance our process won't recognize
that disturbance, and therefore if we see
a change in this flow rate that will then
cause a change in the composition, which isn't
necessarily good when we are trying to control
this outlet composition. So what we can do
is develop a control scheme where we combine
this feedback control with a feedforward controller,
which would be able to better control this
process. Why? Well the reason why is that
if I see a change in the flow what I can do
is have this feedforward controller send a
direct signal to the valve saying that a change
is going to happen, therefore the flowrate
of the manipulated stream can be adjusted
in advance notice of the fact that this flow
is going to be changing. So now what we need
to do is discuss some of the things about
feedforward control. So the first is that
the variable, or variables, being used for
feed forward control must actually vary to
some degree, because if they are not fluctuating
at all there would be no point to have a controller
for it, and if they're only fluctuating in
a small amount it might not be worth it to
introduce additional equipment and what not,
which would introduce costs, and also potentials
for equipment for breaking and what not, so
therefore you want to make sure these variables
actually can vary in some way. The second
part of interest here is that a model must
exist that can relate the feed forward variable
or variables to the controlled variable. So
the beauty of what feedforward control does
is if I know the fact that by changing my
flow rate by 1 gallon per minute of this stream
will cause my composition to go down by 0.02,
let's say mole fraction. What I can do is
I know how to adjust the other stream, the
second stream where the valve is, the manipulated
variable, in order to adjust for that change,
so therefore you must be able to have a model
with some reasonable accuracy, otherwise if
you don't know how the variables are related
you're really not going to gain anything from
this, so therefore having a model is important.
The third obviously is that the variable, or variables
of interest must be measurable. Depending
on the process this may not be as easy as
it sounds, particularly if you're in a process
with high pressure, high temperature, highly
corrosive materials, etc., it may not be as
easy to measure these, but in order for feed
forward control to properly work you need
to actually measure these variables to see
how much they are actually fluctuating from their
assumed steady state value. So this process
here represents a feedback/feedforward controller,
because the part in blue is indicative of
the feed forward process, where as the part of
the control scheme in black represents the
feedback processing. With the feedforward
control scheme the variable of interest, the
controlled variable, is not measured. So therefore
this feedforward controller here is conducting
analysis without knowing whether or not the
composition here is too high, too low, or
exactly where it should be, so that's kind
of an important aspect here. By putting these
together the question is from a block diagram
standpoint what does this look like, and as
a start, we'll start with the part in black
which is the feedback scheme, and really the
feedback scheme is no different than what
we saw before where we had the transmitter,
we'd have a set point with an appropriate
gain. It would then go to the controller,
the valve, the process, which here is the
tank, and then out would come the composition.
So therefore this part has been maintained
by this process. However, the question to
ask ourselves is how does the feedforward
controller fit in, and the way we can gain
some insight on this is kind of looking at
the diagram. What we notice here is that right
before the valve instead of it receiving one
controller signal it now receives two. Both
sum together, and the other one is the feedforward
controller. The feedforward controller gets
the signal from the flow transmitter, which
means the fact the flow transmitter has to
take into account the measurement of the flow
rate of that stream, which we'll call here
F1. By adjusting F1 I am also adjusting the
composition, so therefore there is a second
tank transfer function here that represents
the relationship between F1, the disturbance
variable, and the composition coming out of
the tank, so therefore we have another block
which shows that relationship, so therefore
the composition here changes because of two
things, the change of the flow rate of the
first stream, and the flow rate of the second
stream, which is being manipulated by the
valve. This relationship would come out through
a material balance. Now that we have developed
our feedback and feedforward control scheme
in a block diagram, and therefore we could
conduct a good amount of analysis or put this
into Simulink to kind of see how the processes
act together. So to leave this we'll have
a few parting shots here of well why not have
just a feedforward only control scheme, and
in general feedforward only control schemes
are not very beneficial by themselves, the
major reason why is something that was noted
earlier is that there is no measurement of
the controlled variable, so therefore the
feedforward controller is acting the same
way regardless of whether or not your desired
variable is way too high to little too high,
way too low, a little too low, or just right,
and there's no way for the feedforward part
to know. Therefore feedforward often works
very well with feedback, because feedback
is measuring that value and just making sure
that your scheme is online with what you're
expecting it too be. Another issue with feedforward
only is it's very model dependent, so if the
model that you have is not very strong, or
is not a great fit to the process then that
means that you're going to have some problems
when it comes to actually handling the control
of this process. Again feedback control can
handle that because it measures the actual
value and does comparisons, so therefore if
the model is slightly off it can adjust the
manipulated variable to counteract that. And
a final point, freedforward is not needed
if the disturbance is minimal, or has minimal
effect. You'll just be introducing more softward/hardware
to your process, which could be prone to breaking,
could more trouble than it's worth, particularly
when you have a control scheme already present
from the feedback standpoint. So in this screencast
we took a look at the basic idea of feedforward
control, why a process may benefit from feedforward
control, and what it looks like from a block
diagram standpoint.