Re: Unity gain F/B v. Compensator in F/B

Cameron Dorrough wrote:

"Tim Wescott" <tim@xxxxxxxxxxxxxxxx> wrote in message
news:5sidnVgrY4xoGuzeRVn-hw@xxxxxxxxxxxxxxx

[snipped stuff]

I use SciLab (http://www.scilab.org).  It's a close cousin to Matlab,
with a similar execution window/scripting environment.  In my opinion it
is better in many ways; it's primary drawback for me is that it's
graphical simulator is far more difficult to edit than Simulink.  It is
a _very technical_ environment, but for control system stuff it can't be
beat:

- you can enter a transfer function directly, e.g. H = %s / (%s + 10);
- you can cascade two transfer functions, e.g. G = H * (1 / (%s + 20);
- you can apply feedback, e.g. G_cl = G /. 1;
- you can make Bode, Nyquist, Evans (root-locus) and Nichols plots
  with simple calls
- It will do automatic exact conversions between s-domain to z domain
  (assuming a zero-order hold).
- It supports state-space representations (good for maintaining
  accuracy with high order systems).
- etc.

I like it a LOT; to my knowledge it has all the functionality of the
Matlab with the Control Systems toolbox, it's free and the
English-language newsgroup is quite responsive.  I use it almost
exclusively, with only occasional excursions to MathCad to solve
symbolic problems.



Tim, here's a *really* naive question:  Can you give the curious an example
of the kind of real-world project you might actually use this stuff on??

I'm a "Black Box with A/T" person myself and spend most of my time dreaming
up better ways to connect one black box to another.  I often see adverts for
MathCad and MatLab and lug them straight in the bin - although now I come to
think of it, maybe we did do some MatLab at Uni - but that was an awfully
long time ago.. ;-)

Cameron:-)

_If_ you're working with a system that is well behaved enough that you can make a decent model of it then SciLab (or MatLab) will help you build the model, analyze the model, simulate the model and then make pretty graphs to convince your boss that you know what you're doing. Excel will make pretty graphs, too, but they don't help much later in the project when the system doesn't work the way you thought. MathCad is _much_ better on the analysis side than MatLab -- it's basically a free-form spread *** rather than a command line with a programming language, which is what MatLab or SciLab is.

I think you come from the process control industry, yes? Where the controllers are all accessible, tuning is a normal part of maintenance, and the equipment costs more than the engineering? If so then these tools would only be a real help when you're commissioning a plant or large part of one, particularly if you're not certain if it's actually going to work with the control strategy that you've cooked up.

Where I end up using these tools is for OEM systems where a number of them are going to be built and the customer needs to build controllers to one design and the tuning is fixed. In that case then I need to be able to predict the system performance over manufacturing variations and aging, because a system that isn't up to snuff at build time or after wear isn't "in need of tuning" -- it's "faulty", and if it's faulty because of the tuning then it's a design error on my part.

If you never find yourself staring at Excel thinking "God this is lame" then you don't need MathCad. If you never find yourself writing little programs in C or BASIC to do simulations or analyze a system then you don't need SciLab or MatLab.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com
.