Smith Prediction

[Blackcountryman]

Can anybody point me to information on "Smith Prediction" I'm told that it is a variation on 3 term control that can look forward to what will be happening in the future.

I can find many articles that refer to this but not a publication that gives me the basic information.

As ever I need it yesterday/

 

[ScottyUK]

You might do better to search for "Smith Predictor". I haven't found a good link yet, but I'm looking.


----------------------------------

Sometimes I only open my mouth to swap feet...

 

[Blackcountryman]

ScottyUK,

Tried that, I found some info on the isa.org page but this is mainley papers abbout it's application not the basics.

 

[PNachtwey]

I just googled "Smith Predictor". I got lots of hits that explain the basics. The hard part is finding the model. The Smith predictor is easy. The system identification is not.

 

[MehdiDelrobaei]

Many industrial systems can be modeled using a transfer function with time delay. This time delay causes many problems in analysis of the system. To control a system with time delay, a Mr Smith (in around mid 1970s, I guess) has proposed a compensation technique that adds a feedback loop around the controller to eliminate the delay term in the closed-loop characteristic equation.

Mehdi Delrobaei

 

[dagur777]

try this site

http://controlsweekly.com/controllers.htm#SmithPredictor

Regards,

Dagur

 

[PNachtwey]

good links, dagur77. Since no one feels satisfied I will explain further. Look at this link

http://www.control.lth.se/~fusyntes/lectures/lec34_2004/Presentation_peter_A.pdf

and look at the slide show. There is a slide that is titled control structure. One can see the control output is sent to both the plant and the model G0(s). The model for G0(s) is the model for P(s) WITH OUT THE DELAY exp(-s*T). The PV calculated from the model is then used for feed back and is used for correcting errors in the model. The output from the model is delay by the same exp(-s*T) as the plant dead time T. This delayed PV is subtracted from the current plant value and this difference is added to the feedback. Hopefully there isn't much difference between the model PV with dead time applied and the plant PV. This means the model is good and the controller is mostly using the model feedback as if there is no dead time.