Talk:Robust control
 | This article is rated Start-class on Wikipedia's content assessment scale. It is of interest to the following WikiProjects: | |||||||||||||
| ||||||||||||||
Some remarks to the changes in the last 24h[edit]
It's quite funny: until yesterday noone changed this article for the last 5 months but at the moment one (in this case me) changes the article a number of changes follow. However I do not agree with everything.
- The main topic of robust control I see is robustness against changing of parameters. Although robustness against noise is mentioned at some places I would not mentioned it in the 2nd sentence of this article. But if one thinks, that this is realy necessary i prefer (e.g. parameter variations, noise)
- "Noise" is too loaded of a term, and it causes some unnecessary confusion. That being said, "noise" need not describe a stochastic process. "Noise" describes any variable whose realizations come from some distribution. For example, if I do not know the initial conditions of a system, I can consider them to be a random variable. For example, let's say that I have an open-loop system described as a quadrupole integrator (). To track a reference signal, I could simply feed its fourth derivative into the input. However:
- That configuration assumes that the initial conditions of the integrator are zero.
- If my reference signal is a third order polynomial (or less), it has no fourth derivative. In fact, it can only be tracked by the open-loop system if the initial conditions on the system are set perfectly.
- If there there is any disturbance (or even lay "noise"), the integrator will drift (possibly violently) away from the desired output.
- All of these problems can be fixed by wrapping state feedback around the system (and continue to drive the fourth derivative into the open loop system in a feed forward way). The state-feedback methods make the tracking system robust to all of these things. Robust is robust. It's not simply "parameter" variations (however, because stochastic processes can be parameterized, you can view dealing with them in the same way as dealing with parameters from a compact set). So in principle, it's totally OK to MENTION noise (so long as you're clear what you mean and mention that there are other things too), but because this discussion even came up, the audience of this article is not going to be able to interpret what "noise" really means, and so it should be used sparingly. —TedPavlic (talk) 20:22, 8 May 2009 (UTC)
- I also do not like "Robust methods aim to achieve robust performance and/or stability in the presence of bounded (and often, but not necessarily, small) modelling errors." for two reasons
- Robust control (as I learned for J.Ackermann Robust Control) does not only try to achieve stability but also other aims like dynamic of the closed loop. But this was already mentioned before I started my changes.
- As before I want to state: robust control does not handle with small modelling errors. For small errors standard methods are enough. Remember that we introduced closed loop control to a system because the open loop control is not robust enough against modelling errors. Especially think about the integral-part of the PI-controller.
- Re-read the sentence that you quoted. There is nothing in the sentence that implies that robust methods ONLY handle small variations. That's why the phrase "not necessarily" is used. It's wrong to imply that robust control doesn't have limits. Having a compact set of possible parameter values is required, and for the best results, that set better be small. That doesn't mean it has to be, but your domain of semi-global stability might be tiny. You can't get something for nothing. —TedPavlic (talk) 20:22, 8 May 2009 (UTC)
For this reasons I will change the mentioned lines again. --Hfst (talk) 19:34, 8 May 2009 (UTC) One more thing. I would like to move the example "high gain feedback" to some later place but I am not sure how to combine the two examples. Maybe someone else finds a good way.--Hfst (talk) 19:48, 8 May 2009 (UTC)