Tuning PDF loop (pseudo derivative feedback)

If you really want to make the loop work well, and if servo response (as opposed to just disturbance rejection) is what you're after, then you want to tune the loop for both.

The 'traditional' PID loop basically has the feedback and feedforward terms equal: anything that shows up on the command shows up in the error, which goes through the one-and-only loop filter, and thence to the plant.

Phelan's "huge" breakthrough was to set the proportional feedforward to zero.

What you want to do is to build a loop that lets you adjust the feedforward independently of the feedback. Build a Phelan-esque loop, but inject some of the command into the summing junction just before the plant.

Then tune the loop for good stability and disturbance rejection. Then DON'T TOUCH THAT. Then play with the feedforward gain until you have the step/ramp/whatever response that you like.

There's fancier things you can do, particularly if you know what you're servoing to ahead of time, but you can go a long way just by tuning the feedforward.

Note that this works with derivative control, too: you can put both the proportional and derivative just in the feedback, or you can make them equal in feedback and feedforward, or you can use feedforward that's somewhere in between full and none.

If you go and reinvent sliced bread, then go around insisting to everyone that you have an invention that's the best thing since sliced bread, we'll get tired of you, too.

It's not. It's just that it's not known as "Phelen's Super New Cool Method", or PDF. It's just another well-known (and known before Phelan) way of arranging a loop.

How's that? PID has every step needed to control the loop, more so than what you were referring to earlier..

You can set up a PID to make it work any way you wish.. however, there are differences in how they are implemented depending on the type of output the device has..

A good PID device has analog output and thus the derivative can behave as analog instead of the crappy digital time method used.

Most likely what you are referring to is a basic feed back that involves gain error and integration only, no D equations. Those can be found in the basic closed loop systems like drives for the main speed loop. PID's are an added function for input control between a process variable and control variable from the outside world..

Jamie

He seems to have independently discovered the economy of using phasor algeb ra instead formulating systems of differential equations, computing the LaP lace transform and solving, for the case of sinusoidal steady state computa tions. I'm pretty sure Steinmetz predates him by about 50 years on that one .

Someone posted this link earlier in the thread

It is very good

Read the section about the two ways to create the derivative term, either from the error or from the PV directly.

Mark

There is a information about the algorithm here:-

You might find some clues in those.

Hi Tim, Hey thanks for your other response. So what do people in cotrol theory call it? I found nothing similar in the control guru webpage linked to above. (well I only did a quick browse, looking at signal flow diagrams.) this,

George H.

I don't think it has a name. At least in aerospace and motion control circles, showing it to someone elicits a very blase "oh yea, I do that sometimes, too."

Yep, ZN will give you a nice quarter-critical damped response if your loop is totally linear and time-invariant, but using it where there are interacting loops or other Difficulties isn't recommended. I find that the Louisiana State University approach works well on industrial loops, where some degree of nonlinearity and interaction are almost always present to some degree. It's also just as simple to apply as ZN if you put the equations into a spreadsheet. Sorry, had a quick hunt for links but they're not immediately to hand. Sure to be there somewhere.

and down, measure the peaks and distance between the peaks, and calculate K p, Ki, and Kd.

intrigued by it. There isn't a lot of information out there about it, but nearly everyone that's used it has good things to say.

it slightly simpler. An auto-tune algo like the one I just mentioned would be great, or at the very least, some general guide for manual tuning. A m ethod that doesn't mention poles, zeros, or transforms would be nice becaus e even though I have an EE degree, my strong point isn't controls.

n't know which.

All this tuning stuff is nonsense. You draw a Bode plot of your system and design properly the controller. Tuning is for amateurs.

p and down, measure the peaks and distance between the peaks, and calculate Kp, Ki, and Kd.

m intrigued by it. There isn't a lot of information out there about it, bu t nearly everyone that's used it has good things to say.

s it slightly simpler. An auto-tune algo like the one I just mentioned wou ld be great, or at the very least, some general guide for manual tuning. A method that doesn't mention poles, zeros, or transforms would be nice beca use even though I have an EE degree, my strong point isn't controls.

don't know which.

d design properly the controller. Tuning is for amateurs.

Sadly, there aren't a lot of professionals around with a really good grasp of control theory - and those few that I have met included some that could talk the talk rather better than they could walk the walk.

For the rest of us, tuning is a quick and dirty - but effective - way of ex tracting the information that you'd need design the controller properly in the first place.

And you sound like an academician who's never actually BUILT anything... just paper doodles.

The real world is non-linear. ...Jim Thompson

It also benefits from those local loops we've been talking about. For instance, if you have two loop components that both have 2:1 gain variations over their working range but are faster than the rest, a single large loop has to be slowed down ridiculously to maintain stability. OTOH, if you use local loops, you can make the outer loop something like four times faster, and get much improved transient response.

There are lots of things that are hard to calculate from first principles, e.g. the transient response of a multi-stage TEC.

Cheers

Phil Hobbs

Yep. In my ASIC designs, transistors are dirt cheap, so I really go hog-wild with local loops so that my building blocks approach ideal performance. ...Jim Thompson

up and down, measure the peaks and distance between the peaks, and calculat e Kp, Ki, and Kd.

am intrigued by it. There isn't a lot of information out there about it, b ut nearly everyone that's used it has good things to say.

es it slightly simpler. An auto-tune algo like the one I just mentioned wo uld be great, or at the very least, some general guide for manual tuning. A method that doesn't mention poles, zeros, or transforms would be nice bec ause even though I have an EE degree, my strong point isn't controls.

don't know which.

nd design properly the controller. Tuning is for amateurs.

Actually wrong, it's the academics that perpetuate the PID rubbish. There r ae many problems with this approach. The main problem is for electro-mechan ical systems is that you have no concept of bandwidth. You are measuring ve ry little. Time-domain design is futile. I admit if you are designing say s ome form of chemical plant then that makes sense, since you have little kno wledge of the plant and it is very slow anyway. Look at how a hard disk dri ve is designed and that is how its done. Watch out for the structural reson ance, that is the key point.

I have yet to see a real controls academic that considers PID remotely interesting. Maybe at trade schools, check your matchbook covers.

Best regards, Spehro Pefhany

Well you're talking about the fancy research stuff, but students still need to know the basics too and I gather they do all learn about PID and a litt le of the lag-lead stuff which is far more of use. Trades people aren't muc h use because they don't know what's going on most of the time, though I am sure they could tune a PID. Academics will draw you a root-locus some of t he time of course which is totally useless! You don't need fancy control al gorithms for most small problems but for some (eg control of the Euro-fight er and missile guidance) the modern stuff does get used, so don't knock it. Amateurs are even using Kalman filters in embedded systems nowadays (for i nertial measurement). In the 1970s these were Ph.D topics.

Real men use a Nichol's chart >:-} ...Jim Thompson