Auto Tuning and Why Auto Tuning Doesn't Always Work

http://www.designnews.com/article/511707-Auto_tuning_Hydraulic_Control_Systems.php

The article does show that auto tuning can be done on non-linear systems. In this case it is done on a non-linear Bosch Rexroth valve. The graphs show that it is possible to estimate position far in advance if necessary and there is avaliable processing power to do so. The estimated positions, velocities and accelerations are noise free and you can see by the graphs that the estimated velocities match the actual velocities very well even over a second into the future.

Auto tuning motors is easy.

Peter, have you read anything on fuzzy NN plant matching dynamic gain techniques?

I find it fascinating.

Although it wouldn't be particularly advantageous to the systems you're describing considering there's no call for system emulation from the NN.

But fuzzy gains are interesting all the same.

Honga, I am not impressed with fuzzy logic at all and neither FL or NN will be able to predict as precisely far ahead into the future. I am interested in NN but that will require a lot of processing power that has only recently become available at a reasonable cost. I would like to make a controller one day where you scale the feedback and enter the positive and negative limits and the computer learns and keeps learning as the system changes. I don't think this would ever be a precise as what can be done with exact models, PIDs and feed forwards but not all applications need that precision and the adaptibilty aspect is intriquing. Less tech support for us and faster startups for you. The same basic technique that lets the computer learn is similar to the technique I use to find the coefficients that minimize the error between the actual and estimated values.

" I would like to make a controller one day where you scale the feedback and enter the positive and negative limits and the computer learns and keeps learning as the system changes."

Challenge accepted!

Just in reading the hydraulic application section of the article, this seems to be describing exactly the system I was describing with fuzzy gains. Comments?

A PID based controller that selects gains based on it's current value.

Honga said:

Challenge accepted!

Click to expand...

You are biting off a lot. I have read many articles about hydraulic controls using fuzzy logic. No one has been successful at making a commercial one yet. Also, when the comparisons are done to a normal PID they don't have the PID tune right so the fuzzy logic solution looks better than the PID. When they make the PID look bad then I know they are clueless. On top of that they don't use feed forwards or the second derivative gain so then I am sure they are clueless.

Just in reading the hydraulic application section of the article, this seems to be describing exactly the system I was describing with fuzzy gains. Comments?
A PID based controller that selects gains based on it's current value.

Click to expand...

There have been many people that say they want to implement a fuzzy logic this or that. Most haven't got past the want.

Implementing a FL program is not that hard but teaching to update itself is. A PLC does not have the speed for this or an easy way of programming these kinds of algorithms.

We do our auto tuning on the PC not the motion controller. The motion controller just gathers the data.

Most people can't even chose the right PID to do point to point positioning using a PLC let a lone do all the fancy stuff you are talking about.

Honga, I just came across this.

Honga said:

" I would like to make a controller one day where you scale the feedback and enter the positive and negative limits and the computer learns and keeps learning as the system changes."

Challenge accepted!

Just in reading the hydraulic application section of the article, this seems to be describing exactly the system I was describing with fuzzy gains. Comments?

A PID based controller that selects gains based on it's current value.

Click to expand...

http://khup.com/download/0_keyword-servo-hydraulic-system/analysis-of-a-servo-hydraulic-system.pdf

I know this stuff already but it took years to learn and I had motivation to learn it as I make money selling hydraulic motion controllers.

I still don't like fuzzy logic. Forcing the same users that barely understand PIDs to use Fuzzty Logic is insane UNLESS the system is self teaching but Fuzzy Logic forces the user to setup sets etc. The user must know too much. Neural nets seem better to me but they take so much CPU power that they have been impractical until lately. The back stepping and learning process is the key. If the controller can't learn or adapt then a PID with a good model is just as good. People are relatively comfortable with PID and auto tuning.

The system must be able to tune it self after the scaling is done and the postive and negative limits are set.

Note that the article doesn't really build a controller.