Dear Mr. Skogestad,
My name is Thomas Petsopoulos. I am an aerospace engineer, with
Standard Missile Company. As the name implies, our company makes a
missile called Standard Missile. Despite its boring name, Standard
Missile is actually a high-performance tail controlled ship defense
missile (though it is evolving different roles).
I am currently working on an Advanced Technology Demonstration
Program, in which we are investigating the use of new robust
multivariable control system synthesis techniques to redesign the
autopilot/guidance system. One of the goals of the project is to
demonstrate improved missile time constant by "simply" replacing the
current autopilot, with a "high-performance" multivariable autopilot.
My specific task is to use the mixed-sensitivity H-infinity design
methodology to redesign the autopilot (though there are several other
"new" design techniques that are also being investigated). I
purchased your book a month ago, and I have found it very useful.
The plant used in the synthesis of the autopilot has been limited to
the rigid body and actuator dynamics, in order to reduce the number of
states and the size of the state-space controller matrices that must
be gain scheduled (read CPU throughput limitations). However, I have
been analyzing this "lower" order autopilot using a much more detailed
and sophisticated linear plant model. The analysis code I have
written is digital, and it includes transport delays in computing the
fin-servo voltages, high-order structural modes, structural filtering,
IMU sensor dynamics, etc.
I have slowly watched my wonderful gain and phase margins (determined
by opening one loop with the others closed, both at the servos and
after the sensors) degrade with each level of "real-world" realism
that I have added. We can't fly at White Sands Missile Range unless
we have 4dB/20deg in all channels (and they don't believe in "mu").
In addition, we haven't even looked at the nonlinear
simulation/hardware-in-the-loop response yet!
Well, the next step is to develop an integrator anti-windup algorithm.
For some reason, none of the texts on multivariable design address
integrator anti-windup. Your book was the only one with a page devoted
to anti-windup, but it was targeted towards a 2-DOF loop shaping
Hinfinity design, like the Glover McFarlane technique. Has anyone
ever looked at integrator anti-windup techniques for the
mixed-sensitivity Hinfinty controllers, and if so who, what, where,
when and how?
Any direction you could give me on this matter would be greatly
appreciated. And thank you very much for your time. I have no money
to give you, I'm just asking for a favor.
Tom
P.S. By the way, I thought that your book was the most useful text I
could find on multivariable control. However, I would like to make a
suggestion. If this "theory" is going to be of any use, someone has
to write a text that actually takes one of these multivariable designs
to completion. In that I mean that I would like to see the actual
digital controller, saturation/anti-windup logic, nonlinear gain
scheduling, sensor noise/actuator heating results, structural filters
(high frequency design), and any of the special problems/solutions
that are needed to get one of these high-order Hinfinity controllers
to work.
Where are the experimental results? The aerodynamicists have always
compared to the wind-tunnel data, the finite element guys always
compare to structural/vibrational tests, but the control theorists are
somehow never held accountable for their theories!
Let's see an actual hardware comparison of an optimal robust
multivariable design and a good classical Bode design. Is there any
real advantage of this new robust multivariable theory? If our stuff
wasn't classified, I would write the book (of course, I should
probably finish my part-time Ph.D. first, so I am at your guys level
mathematically!) You don't need to apply the theory to fancy military
hardware ... just point a flexible beam, or something simple.
Nonlinear, adaptive, multivariable missile autopilots have been around
for decades. The guys who designed them used classical sequential
loop closure and gain scheduling. They work well. These guys are now
managers, and they need good reason to use other methods, and rightly
so.
You have an interesting quote in the beginning of your book about
"playing" around. I would read the Preface to the Second Edition of
Garnell's book on "Guided Weapon Control Systems". In addition, look
at the pictures of the HARDWARE. Garnell's book may be "simple-minded
classical/analog design", but this is reality.
Well that's all I'm going to say, and I've probably said too much
already. Thanks for listening.
Received on Tue Apr 29 16:35:09 1997
This archive was generated by hypermail 2.1.8 : Tue Jan 20 2004 - 09:32:17 CET