PID Loop Learning

[Peter Nachtwey]

i do not agree with you peter, although it is often thought to be that way.
I see Fuzzy more as a expert system, so with the knowledge of an EXPERT you make a control system.

There is nothing expert about Fuzzy Logic. It is simply trial and error.

I have a demo with 3 distance sensors in a racecar taking a course.

I bet I could do a better job with PID. I may have to augment the PID but the key thing is knowing the plant values.

I think fuzzy is often misunderstood, i am calling it more a expert system.
PID is so easy with autotune anybody can do it. (but only good for one situation)

So when you are done, how can you call yourself and expert if you don't have a model of the system you are controlling?

The way to do this is to use the chain rule, remember calculus? You don't find this in a book anywhere. The trick when using a PID on conveyor or motion applications is that you don't execute the PID as a function of time but as a function of distance moved. Then that output must be scaled by how fast the conveyor or car is moving. If the car isn't moving forward you don't want the proportional and integrator gain to cause the care to turn a sharply when it can't. Also you don't want the integrator gain to wind up when the car is not moving. The integrator and derivative terms must even updated as a function of distance traveled.

All systems can be modeled using poles and zeros but mostly poles. I know this and add PID gains or even second derivative gains to place the closed poles exactly where I want them so I get my desired response. In the case of cars changing lanes I can calculate the gains so the car will successfully change lanes after traveling so many feet or meters.

Because I know how to model systems and do system identification is it easy for me to calculate gains.

Fuzzy Logic doesn't require a model but neither does it allow you to get exactly the results you want because there is no way to do the math. There is just trial and error. Some of the better fuzzy logic software will try to optimize the response by trial and error but a system identification or optimizing algorithm is used. The result may work but there still isn't a plant model. If you have plant model there is no need for fuzzy logic. The gains can control method can be calculated exactly.

If I were a professor teaching control theory I would insist that the students be able to write the transfer function for the plant.

 

[RET]

Just remember, half your constants when tuning.

 

[Peter Nachtwey]

Just remember, half your constants when tuning.

Why?

 

[kamenges]

Because the system isn't mathematically modeled and you can't calculate the correct gains, of course.

 

[Peter Nachtwey]

Because the system isn't mathematically modeled and you can't calculate the correct gains, of course.

But half of what?

If your plant is a FOPDT or SOPDT I still have a Scilab program that will calculate gains. I still have the hotrod.txt file provided by Ron Beaufort.
http://www.deltamotion.com/peter/Scilab/AutoTune/

There are three lines. One is the PV from Ron Beaufort's hotrod trainer. The other 2 are the EV, estimated values, calculated by the AutoTune.sci program. The two estimated responses match the actual responses closes.

Someone on LinkedIn used my AutoTune.sci program. It worked well so he added a HMI or front end to the program.

 

[shooter]

Lets make a new subject on fuzzy as OP is struggling with PID.