What is model-based control? 1

[PerceptualRobots]

How is model-based control defined?

I have some familiarity with PID-type controllers, which are not model-based, I assume, so what is model-based control?

Regards,
Rupert
perceptualrobots.com

 

[xnuke]

See Link.

xnuke
"Live and act within the limit of your knowledge and keep expanding it to the limit of your life." Ayn Rand, Atlas Shrugged.
Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[PerceptualRobots]

OMG! A way you can search the internet, if only I had known!

A wonder what the point of a forum is if you are just going to refer to google?

Regards,
Rupert
perceptualrobots.com

 

[EmmanuelTop]

It means you haven't done your part of the work, because:

1) You asked a very general question;
2) You haven't stated what kind of research on the subject you did before posting on the forum, and what still remains unknown to you after the research;

The question you raised is like having an engineer asking "what is a centrifugal compressor" and for proper response one would have to write a book (or a chapter at least), because there are so many things to say if we start from basic stuff. Purely based on your question, we have no idea where are you coming from.

Dejan IVANOVIC
Process Engineer, MSChE

 

[PerceptualRobots]

Well, if someone doesn't know what MBC is, then how would they know that? Besides if the topic is as nebulous as you say then google isn't going to be much help.

Anyway to have a discussion with people on here it is necessary to first find out what their understanding of the topic is, rather than what google brings up, which was the point of my question. Once that is established then I will know what questions to ask to progress the discussion further.

So, let me rephrase the question.

What do you understand is meant by model-based control?

Regards,
Rupert
perceptualrobots.com

 

[IRstuff]

Seems pretty clear to me:

https://en.wikipedia.org/wiki/Motor_control#Model_Based_Control_Strategies

MBC is contrasted with more heuristic approaches. MBC is, by definition, control of a system using its mathematical representation in the form of differential equations, etc. Hence, PID is MBC, since a PID control implements P, I, and D in equation form.

TTFN
I can do absolutely anything. I'm an expert!
homework forum: //

http://www.engineering.com/AskForum/aff/32.aspx

faq731-376 forum1529

 

[PerceptualRobots]

> Hence, PID is MBC

My understanding is that PID is not MBC. As you link says MBC "rely on accurate internal models of the environment, constructed from a combination of perceptual information and prior knowledge", in what way does PID incorporate internal models of the environment, or prior knowledge?

In MBC what type of information, or prior knowledge, comprise models? For example, what would be required for a cruise controller?

Regards,
Rupert
perceptualrobots.com

 

[IRstuff]

PID is implemented in either software or hardware as stated, tied directly into the plant. That makes it an MBC, despite what some people assert. MBC applied to PID is really PID 2.0, which accounts for dead time, which is a plant behavior, adds adaptive control, and adds feedforward.

TTFN
I can do absolutely anything. I'm an expert!
homework forum: //

http://www.engineering.com/AskForum/aff/32.aspx

faq731-376 forum1529

 

[PerceptualRobots]

What, then, are "models" in PID? That is, in what way are internal models of the environment incorporated?

Likewise, what form do models take in MBC (that are different to PID)?

Regards,
Rupert
perceptualrobots.com

 

[roydm]

Model based control is an excuse to sit in front of a computer screen.
Meanwhile the instrument tech wanders out and fixes the problem LOL

 

[itsmoked]

MBC is used in more complicated cases than PID. For instance the Aleyska Pipeline is Model Based Controlled. They have crazy-ass big models of the pipeline that they use to control the pipeline based on inputs from various sensors. Usually in these cases they have some form of diaper-code to detect things going badly haywire to try to herd the whole thing back into complying with the 'model' perhaps at the momentary expense of efficiency or delivery schedule.

I worked with one of the models that dictated where the product was supposed to be and hence flows based on the what flow sensors where providing. If they had a leak somewhere then after a while the sensors would be disagreeing with the model enough to cause a general stirring of the anthill.

Keith Cress
kcress -

http://www.flaminsystems.com

 

[roydm]

Keith,
Thanks for posting that example, my experience is more with in plant controls where some young wizz kid wants to model a few simple loops.
I worked on a 90 km water line once, the pressure surges in that were downright frightening, several minutes delay between making a change and the pressure wave arriving.
Roy

 

[itsmoked]

Interesting Roy. You can imagine the Aleyeska pipeline being 48" diameter and 1300km long with slackline areas being a bit rough to model.

Keith Cress
kcress -

http://www.flaminsystems.com

 

[PerceptualRobots]

Thanks for the example, though I couldn't see the details of it. Could someone point me to a simple example of MBC that uses an internal model? I am working on a control system architecture that doesn't require internal models and I'd like something to use as a comparative. I haven't been able to find anything suitable from a search yet.

Regards,
Rupert
perceptualrobots.com

 

[greg2835]

I will try to put this into the simplest terms, and the other posts have stated this but hopefully this clears things up.

Model Based Control is exactly what is sounds like. Designing a control system based on a model of the system. So if you have a transfer function of the system and design a controller (PID, lead/lag, just K, etc) that is model based control. For more complex systems you can have several equations (a model) that represent how a system behaves to various inputs and design a controller (adaptive, robust, etc).

Now, Non Model Based Control is when you design a controller without knowing what the model of the system (plant) is. An easy example of this is an automatic motion solar receiver. Let's say the goal is to maintain a maximum UV input to the receiver which we can measure. In order to do this I am not going to determine a model for how the UV changes as the sun travels across the sky, I just want the receiver to get the most UV it can. So my controller logic would be something like this:

- Move a little bit in one direction
- Did the UV input increase?
- Yes, so move a little more in that direction
- Did the UV input increase?
- Yes, so move a little more in that direction
- Did the UV input increase?
- No, so move a little bit in the opposite direction

This algorithm basically is looking for the peaks of the UV and trying to maintain it, without using a model of the system.

Hope this helps!

 

[PNachtwey]

There seems to be some confusion here:
I wouldn't say a control system is using model based control unless it was really using a model. A simple example is a Smith Predictor.
If a model is used to calculate gains then I would call that model based design.
A PID is not a model. A PID cannot be use to predict what will happen if I apply a control signal to it. The PID gains are simple parameters that move the closed loop poles around.
Feed forwards are open loop terms and yet they could qualify for model based control as feed forward gains are calculated by inverting the open loop transfer function.
Still I would NOT call using feed forwards model based control, but it comes a lot closer than a PID. At least with feed forwards I can predict the response using the set point and its derivatives.

Model based control is when the control's output goes to the physical system and to a model like what happens in the Smith Predictor. The response of the model is compared with the response of the actual system and any error between them is used to correct the current state.
Here is an example. I am simulating a SOPDT system. The control output from the sliding mode control, SMC, is put in a delay queue to simulate the dead time. I can make the dead time as short or long as I want. In this example the dead time is relatively long compared to the plant dead time.
To make things more interesting I use a random function to simulate the fact that I may not know the plant parameters exactly. It is too easy if one does. Each time I hit ctrl-F9 a new simulation is created with different random values. I know a few of you are Mathcad users and will immediately recognize that I am using Mathcad.

http://deltamotion.com/peter/Videos/SOPDT Sliding Mode Control with Smith Predictor.mp4

Peter Nachtwey
Delta Computer Systems

http://www.deltamotion.com

 

[itsmoked]

Totally agree with you Peter.

Keith Cress
kcress -

http://www.flaminsystems.com