Minor Loop Control questions

"Peter Nachtwey" schrieb im Newsbeitrag news:sOadnRRUbtIAervbnZ2dnUVZ snipped-for-privacy@comcast.com...

If you tune power stations you know by experience that this type of steam generator and turbine can be power-incread 10%/minute. Your employer sold for millions of dollars the control sytem that garantees 10%/minutes. The time delays are minutes, not seconds.

u = power target x = power (PV)

On a first try, u is limited to 1%/minute. If all is going smoothly you go to 2%/minute, and so on.

The operator would never have a change to move the plant with a step signal. There will always be a target speed limiter inside the control system.

If finaly power change of 10%/minute is proved and accepted by the staff of the power station my job was done.

What I was basically trying to do you find in the diagram:

formatting link
on bottom

V(theoretical) = high (or whatever is possible) Time-compensated(theoretical) = 100% (or whatever is possible)

In the times I tuned power stations with try-and-error method I could reach

10%/minute 'without destroying' power stations.

That last diagram you must realy understand: x ~ u c can be 1

This gives you 'linearized' output.

V is nonlinear (valves, pumps, etc.). The integral part of PIDs is worsening time behavior of the loop. You must therefore minimize the integral impact. All these thoughts apply also to your system. In your system I've chosen a filter for example, not a speed limiter.

Peter, you have modified my idea to feedback the total process transfer function in a Matlab test. You should try using feedback the total process transfer function. I have the target signal damped by a filter with T=0.25.

If it works you have the advantage that you or anybody else can change from try-and-error methods to calulate the system in less time, having only one parameter to adjust (filter).

To make it clear: Your result you presented was ok for me. I was just trying to calculate a similar result, and if possible do it slightly better, have better influence to control behavior.

What I'm up to is to find also process transfer functions by good approximation methods and calculate the whole stuff. Calculation is better than just trying.

General note: If PT1 can be controlled by PD why not PT3 by PD3?

Reply to
JCH
Loading thread data ...

...

...

Success can be its own justification; I don't question your successes. You must understand, though, that if the system cum controller model predicts instantaneous change in the output, it must be deficient in at least one respect, making all of its predictions suspect. A model is only a model, and if one needed to be exact in order to be useful, none would serve. That said, a model with serious departures from even potential reality doesn't deserve much confidence.

Jerry

Reply to
Jerry Avins

Correct.

My system is G1->G2->G3 with H feedbackbetween G2 out and G2 in, and unity feedback G3 out to G1 in. My readers is so hosed with fonts, I couldn't make much sense out of Tim's blk diagrams...

Thanks,

Bo

Reply to
Bo

Hopefully you have turned in your home work by now. Below is a link to my interpretation of your home work. ftp://ftp.deltacompsys.com/public/NG/Homework-BO.pdf Did I interpret your system correctly? I verified what you claimed about the velocity constant of the original system being 1000 so I figured I couldn't be too far off. It is hard to make sense out of block diagrams written with a text editor. I am not an artist so I didn't take the time to draw a block diagram of my interpretation of your problem but you should be able to tell from the transfer function. Page 5 is where I modified the minor loop to change the velocity constant.

Peter Nachtwey

Reply to
Peter Nachtwey

Modeling in this context means doing mathematically. The results depend on input data accuracy. If input approximation is 97% accurate then you have a good model.

Let's assume Peter's model is of that quality and there is no further significant loss by using appropiate equipment then it would work very well.

I guess that he proved it.

Reply to
JCH

Unless I looked at the wrong result, Peter's model predicts no delay between command and response, and allows step changes in the response. The mathematics may be good, but there can be nothing in reality that it models correctly, and precious little that it models adequately.

Jerry

Reply to
Jerry Avins

=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF= =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF

I think you have us confused. It is JCH that has no delay between the command and the response. It takes my example, in the iterative tuning thread about, 100 milliseconds to reach the first set point.

Peter Nachtwey

Reply to
Peter Nachtwey

...

You're right; I do. My comments applied to the result of his model.

I apologize for slandering you. :-)

Jerry

Reply to
Jerry Avins

I think you have us confused. It is JCH that has no delay between the command and the response. It takes my example, in the iterative tuning thread about, 100 milliseconds to reach the first set point.

I have delay between manual step and target value! This will reduce forces (acceleration).

Have a look at Page 1 and 2:

formatting link
g ~ 10 m/s^2 = 10000 mm/s^2

Force F = m * a a = acceleration m = mass

Reply to
JCH

???????????????????????????????????????????????????????????????????????

Well then, I'm completely confused and slandered everyone. Wjose system gain was k/(k+1) or some such?

Jerry

Reply to
Jerry Avins

=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF= =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF- Hide qu= oted text -

It is hard to keep tract of what JCH is doing. I took me a couple of posts to figure it out.

See this link

formatting link
You can see that the actual ( magenta v2 ) follows the target ( red u ) almost exactly. JCH generates a filtered target position using a low pass filter. The transfer function is v2/u. JCH justs inverts the F1(s) ( the plant ) to get F2(s) ( the controller ) so the product is always 1. Now one can simplify the transfer function to 1/(1+1/k). If all the terms are mulitplied by k the transfer function is k/(k

+1). This implies there is no lag time so a step change in the target position would cause a step change in the actual position. This is not possible without having a way of adding and removing energy instantly.

In JCH's last post he has done something different but it isn't clear what. At least it is more realistic.

I think JCH should look at this. Deadbeat and Dahlin controllers are somewhat related to what JCH is trying to do.

formatting link
Peter Nachtwey

Reply to
Peter Nachtwey

It is hard to keep tract of what JCH is doing. I took me a couple of posts to figure it out.

See this link

formatting link
You can see that the actual ( magenta v2 ) follows the target ( red u ) almost exactly. JCH generates a filtered target position using a low pass filter. The transfer function is v2/u. JCH justs inverts the F1(s) ( the plant ) to get F2(s) ( the controller ) so the product is always 1. Now one can simplify the transfer function to 1/(1+1/k). If all the terms are mulitplied by k the transfer function is k/(k

+1). This implies there is no lag time so a step change in the target position would cause a step change in the actual position. This is not possible without having a way of adding and removing energy instantly.

Jan:

Peter, my last posting shows: Velocities, acceleration. Your actuator should be able to accelerate the mass like you can accelerate your car. The acceleration graph may look suspicious, but it is not.

Example: Starting from 0 to 1 m in 1 s the velocity after 1 s is 1 m/s. And acceleration is 1 m/s^2 immediatly (looks like a step). You need 1 N if mass is 1 kg.

I showed the acceleration in:

formatting link
That is all realistic. If you have mass to move you can calculate the maximum force F_max=m*a_max) looking up acceleration from the graph. If you haven't the necessary force built in your system will fail working properly.

The velocities look also plausible. That means realistic. Why should it be impossible to move mass like that if the force is available.

Maybe you show us your velocity and acceleration. These are the first and second derivatives. I guess MATLAB can do that.

Notes:

1) If you have no process and no controller you can move the mass directly as shown with u, v2, v2', and v2''. (u = v2) 2) In Page 2 I used a more smoothing filter: T^2*u'' + 2*d*T*u' + u = w, d=1, T=0.055. 3) E.g., if demands are 'extremly high' (d=2, T=0.002) then you must have the force for accelerating ~ +/-25g. That shows technical limits.

See also

formatting link

Reply to
JCH

Dnia 15-04-2007 o 17:16:23 bo napisa³(a):

(...)

Could you exlpain what is minor loop and how you converted your controller? Could you show block diagram or give a link to some pdf's or something. Try to use AACircuit v1.28.6 beta 04/19/05

formatting link
proposed by Tim Wescott. Try to configure Forte Agent to look diagrams nice an proper if they don't.

(...)

My definition of plant is diffrent. Plant is something you can't split. Unless you have 3 or four plants in your Gp and among others there is one with Gxp=1/(1+s/100) transfer function. Yo cannot connect anything to any inner term of your plant. It's an abstraction. It could be radial acceleration of some wheel or third derivative of ions concentration or even something stranger and impossible to describe. This is just abstractive model. You can reach only input and output of it. That is what the transfer function represents, input-output relation.

Reply to
Mikolaj

That is limiting, but lets work with that. What about a hydraulic cylinder and servo valve with LVDT feedback. The servo valve LVDT is the inner loop feed back and actuator position feedback is the outer loop. Therefore by your definition the valve is one 'plant' and the actuator is another 'plant' . OK. As long as it works out the same mathematically. Motion controllers often have an inner velocity loop and an outer position loop. All of this is on one card or module. What about this? Adding an accelerometer can also provide feedback to an inner loop.

It is obvious that you can, but you are right. In some cases it is just an abstraction. Some motion controllers use the encoder feedback as the position feedback and the velocity feedback. It is just a couple lines of code. I think you are just hung up on the idea that the valve and the actuator can be consider just one 'plant'.

I agree that one shouldn't states that can't be measured as feedback in an inner loop. It wouldn't make sense to do so. But what if you can measure the acceleration with an accelerometer as I said above?

Peter Nachtwey

Reply to
Peter Nachtwey

In practice when you see a "plant" on a block diagram it means little more than what the guy drawing the diagram wanted to call a "plant". If I'm building servo-controlled valves, then my "plant" is your "actuator". If you're buying my valves to build diesel engines, then your "plant" may be a crane-building company's "actuator".

So, in general, you can't say that a block labeled "plant" is irreducible, and you can't say that a reducible block that's labeled "plant" is done so incorrectly.

I have absolutely no argument to the idea that you can't reach inside of an irreducible block and pluck out an inner state, however.

Reply to
Tim Wescott

Dnia 23-04-2007 o 18:34:53 Tim Wescott napisa³(a):

(...)

I agree, but there are plants that cannot be describe analyticaly. (I worked with thermal process that is known only experimentally and I can't connect anything inside it's transfer function) I agree that simple mechanism can be described exactly (more or less we know there is a mass and a viscosity friction and a spring).

That is not the point. When talking about homeworks we use "blocks" but we must remember of reality. We should specify what we can and can not do with our blocks.

I saw models of hydraulic valves but who in practice analyses it's gap shape influence on some transfer function (well, maybe valve inventors )? Can you get inside the valve? We know it's block model. That's why we need observers. To observe something that maybe has interpretation but not necessarily.

Reply to
Mikolaj

Yes.. You have it. I should get my HW back this evening...so we'll see.

I think you have Ti right on page 5--but your overall transfer function (I think) shows the series compensator and the minor loop compensator in the system--instead of just the minor. Take out the series compensation and I'm pretty sure the kv will then no longer be 1000.

Nice document-- what did you use to create it? Matlab?

Thanks for taking so much time--that was way more effort than I was asking for--and I sincerely appreciate it.

Best regards,

Bo

Reply to
Bo

I used Mathcad generate the data pdf995 to generate the .pdf file. Mathcad can generate html files but Mathcad is so screwed up that all the links to the pictures use Microsoft back slashes instead of UNIX forward slashes so the html files don't work well with non Microsoft browsers. I have a love-hate relationship with Mathcad. It does some things extremely well ( symbolic math ) and other things are totally brain dead. Still it is a good deal if you can get a student version. I don't think I could have learned what I have without Mathcad or some other symbolic math package. Matlab and Scilab are good for getting answers. Mathcad is good for figuring out how that answer are derived symbolically.

Peter Nachtwey

Reply to
Peter Nachtwey

Dnia 16-04-2007 o 15:30:52 Bo napisa³(a):

(...)

(...)

Don't worry, me neither, so you are not alone :). [Jan, what the ... are you talking about?]

Kv=lim s->0 (OpenLoopTF) So, it depends on your TF's (transfer functions) if inner loop can influent on Kv. If they are zero class type or something A/sB (A,B 's' polynomials).

The structure is crucial. This the first thing we have to determine before any calculation. If you change structure, you change everything.

Reply to
Mikolaj

"Mikolaj" schrieb im Newsbeitrag news:op.tq8no7ej3r65ff@xmasnew...

Sorry, first I didn't recognize that it was homework. In real life I would have a solution for

similar to I discussed with Peter's process transfer function.

I couldn't recognize a minor loop!

I'm talking about 100% time-compensation using PD^3 controller with loop gain as high as possible avoiding or minimizing controller integral function.

Using Peter's process transfer function he posted:

See revised version:

formatting link
Page 1: Math model: Fig. 1 Loop model: Fig. 2 The behavoir would then be like an amplifier, Fig. 3.

Page 2: If setting a benchmark test w the result could be like: Page 2

That's what I had done if it were not homework but real work to be done.

Reply to
JCH

PolyTech Forum website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.