Advanced Control: Rant!

AustralIan said:

Thanks for your detailed reply Peter. A few things clicked for me here, particularly about lambda tuning, and how this is really just pole placement.

Multiple loops interacting- I didn't mean cascade loops, I meant.. Best example is two separate steam valves for separate parts of the process off of one steam header.

Click to expand...

I have done similar projects with hydraulic valves. Obviously you must have an algorithm for the total flow as a function of the output to two valves.

If one opens, the steam through the other valve will reduce, causing it to open until the steam generator picks up its pace. This can cause the loops to oscillate, particularly if you tuned the loops one at a time

Click to expand...

That is OK but there must be some master control loop controlling the minor control loops so the output is smooth.

for as quick as a response as you can get, and for long dead times. How do you avoid and/or fix these interactions between loops?

Click to expand...

Dead times are a separate issue. See my Smith Predictor videos or PDFs.

The blending of outputs requires a blending algorithm to make them look like one.

Moving average + dead time: I was not proposing this as a control method but describing the process. Ie. the process TF is, if I am correct, K_p (e^-s - e^-4s)/s. My first instinct was model it as fopdt, and use a tuning rule, but the d term does not like that kick at the end.

Click to expand...

Kp/s is a system with no dead time. Why do you have two dead time terms?

The tank gain changes are "easy"* in that you have enough information to calculate them precisely. What about things like "how will this behave in winter?"

Click to expand...

What do you mean? The tank dimensions and orifices don' t change in winter.
However, hydraulic oil is thicker ( higher viscosity ). Saw mills are usually exposed to the out door temperatures. The usual procedure is to cycle the hydraulics back and forth to warm up the oil to normal operating temperature before production starts.

It is possible to tune for cold oil but no one does that.

or "when another machine is added to the steam header." Or "it is windy." I guess what I am really after is tuning for disturbance rejection by identifying expected limits of disturbances, and choosing both a controller and tuning that are robust enough to handle the expected range of disturbances.

Click to expand...

I need an example with data but it is easy to find one set of robust tuning parameter but they will not be optimal over a range of disturbances and conditions.

Some of those will be unmeasurable changes to plant time constants, plant dead times, plant gains, or just random changes to your PV. A lot of tuning demonstrations only look at the process in terms of controller output to process PV transfer function as designed or as measured.

Click to expand...

In saw mill applications we don't know the size( mass ) of the log. We tell people to tune for the average mass. Feed forwards make a good estimate for the control output for the average mass of a log. If all the logs were the same there would be no need for a PID. However, masses change so the PID corrects for errors but the errors are usually small. This is much better than relying on the PID to provide all the output.

And lastly, I will think more on the instantaneous time constant theory, and its implications.

Thanks again for your dedication to the betterment of control engineers.

Click to expand...

The implications are the plant model is always changing. When your eyes are open and you go beyond what the college professors have told and what you have read in the many control books you see the light. I now understand control thoroughly but if you look at my posts about level control on this forum I usually recommend a simple proportional band unless the level must be controlled precisely.

In the real world, the people on this forum can get by tweaking gains an looking at the response. They are not designers. I realize that. My rant is about college professors that are clueless. For instance. In the two tank level control problem, how high must the first tank be? If the tank isn't high enough it will over flow before there is enough head/pressure to keep the level of the second tank constant. If the first tank is not high enough there is NOTHING the PCL programmer can do to control the level in the second tank.

My rant is about college professors that are not teaching what engineers need to know about simulation and design. Designers must be a cut above. If they aren't then the people that are forced to tune their kludges will be cursing them.

Hi Peter. I think you need a video with no graphs or text that just spells out what someone should learn to be a great controls engineer.

Also, how do you think universities and academia can solve the problem of clueless controls professors? Have you thought about the root cause of the clueless professors being prevelant in academia? Did you ask five whys to get there?

Lastly, I would like to flesh out my moving average + dead time plant a bit more for further discussion perhaps in a new thread. I think the TF is correct:
KP/s is an integrating process
The step response to e^-s - e^-4s is 1 between t=1 and t=4

AustralIan said:

Lastly, I would like to flesh out my moving average + dead time plant a bit more for further discussion perhaps in a new thread. I think the TF is correct:
KP/s is an integrating process
The step response to e^-s - e^-4s is 1 between t=1 and t=4

Click to expand...

This really doesn't make sense. Basically what you have is two Heavyside step functions in the time domain. If you only consider the time between 1 and 4 seconds the second Heavyside step function is NA.
I can use my Mathcad to apply a step function to (Kp/s)*(e^-s-e^-4*s). I don't see where a moving average is going to be applied except that is can smooth out the response between 1 and 4 seconds. Why? What are you really trying to do?

I don't see how e^-s-e^-4*s can be a response due to a integrating process without dead time. Kp/s doesn't have a dead time so where does it come from?

This really doesn't make much sense.

Peter,
I'd like to see a video on Sliding Mode Control (SMC). There was a thread on this a while back and I employed the code successfully but the gain/time constant calculations were a puzzler to me. Thanks!

Just like I have a phone that has a great calculator..... to replace my slide rule...or manual math calculations
I buy instruments with AutoTune to figure out my PID.
Azbil SDC35 or SDC36, as an example, for a near perfect and quick autotune, and it supports up to 8 PIDS.

RonJohn said:

Peter,
I'd like to see a video on Sliding Mode Control (SMC). There was a thread on this a while back and I employed the code successfully but the gain/time constant calculations were a puzzler to me. Thanks!

Click to expand...

I have a Sliding Mode Control with Smith Predictor on my Peter Ponders PID YouTube channel. I really need to make another video with just the Sliding Mode Control without the Smith Predictor. Then I can add simple improvements incrementally.

Sliding Mode Control is great for applications that are controlling on-off devices like solid state relays but not so good if there is dead time. For temperature based systems the Smith Predictor should be implemented.

If dead time is 0 or negligibly small SMC is a great way to control systems that don't need to be extremely precise but must be robust.

SMC is different from normal control loop control. Normal closed loop control is all about picking gain that place poles that provide the desired response.
When using SMC the engineer picks coefficients that result in the desired response directly.

Thanks Peter! I look forward to checking out the current & future video.