It really isn't a case of "better" or "worse" response. In many cases,
normal PID controllers are perfectly adequate. These include the "single
input - single output" controls.
The fuzzy set controls I worked with dealt with multiple input, multiple
output systems. In these cases, it isn't unusual to have actions which are
based not only on the magnitude of deviation from setpoint, but the
direction of deviation (in several dimensions). PID, without a bit of
work, will have one action for a high process value (eg, open cooling water
valve), and its opposite for a low process value (close cooling water
valve). Fuzzy sets can have one strategy for high values (open cooling
water valve, reduce reactant feed rate) and another for a low process value
(add reactant), and another for a very low process value (add steam, turn
Have I obfuscated the difference enough?
email@example.com (Hesham Elhadad) wrote in message
Siemens have for their S7 series of PLCs fuzzy and neurofuzzy control.
For 95% of applications PIDs are perfect and for the remaining 5%
about 3% are adaquete.
Fuzzy is excellent for capturing linguistic knowledge such as "if
getting to medium hot very fast then reduce heat input to lower and
turn fan on low" this info may not be an the form of a known model or
Neurofuzzy helps you tune those rules without actualy getting and
expereienced plant opperators knowledge into human language: you just
let the neruofuzzy learn as it observes the actions.
From my understanding fuzzy is usefull in improving and optimising
plant which is poorly understood becuase it is non linear and multiple
inputs and outputs and hasn't been or can't succesfully be modelled or
measured. It is a way of capturing human knowledge and implementing
it into control.
Siemens have some example projects: I think concrete plants are one.
In some cases a system that requires two PID loops and thus two
sensors eg voltage and current in some variable speed drives these can
be reduced to only one fuzzy loop with only one voltage sensor thus
saving some cost and not loosing performance.
They've been used to estimate whether a car driver is trying to save
fuel, trying to drive fast or trying to drive on slippery snow and
then adjust accordingly. I think VW Golfs have had these for years.
Don't use it as a substitute for lack of knowledge of PIDs.
Actually, I've found fuzzy controllers to be quite stable, often more so
than PID controllers. This is probably because they are able to handle
coordinated responses of many variables while PID cannot.
Thanks for all of you Guys
I think the picture becomes more clear for my eyes now. It is clear
that the PID still has the power as a perfect controller over the new
technologies which is still under research.
Actually, I have studied the fuzzy logic for two years but I have not
supervised it working in nature.
In my working field (DCS) almost of the control is based on the
traditional (PI) control algorithm only.
I have never worked on the siemens PLC before, and hope to do one day.
Thanks for you again,
Look forward to hearing from you soon.
PID control is best for siso systems which are well defined with a single
control rule or a cascaded set of rules.
Fuzzy controls are best for mimo systems which are poorly defined (or
understood), have multiple control rules, show some nonlinearity, or require
disparate actions in different areas of the operating plane. Generally,
I've seen fuzzy sets being used in supervisory control systems, with PID
controllers running the individual regulatory loops. That combination
allows both systems to be used to their greatest advantage.
As far as perfect control, I'm not so sure. I've dealt with a lot of
systems which work fine on PID. I've dealt with a few which did ok on PID,
but much better on a fuzzy controller. I've seen a couple in which fuzzy
sets didn't work (PID wouldn't either). In those cases, there was an
underlying engineering problem with the unit operation.
Each type of control has its place. DCS systems tend to only have the PID
algorithms on them because they are well known, and expected. They are also
derivatives of the old pneumatic controls. Fuzzy sets came much later, and
DCS vendors don't seem to be too quick to put new methods in. If they put
in fuzzy sets, they would have to put in DMC, eDMC, neural networks, lambda
tuning, process bots, etc, and they don't want to go through the effort. It
is quite frustrating to see a system which would benefit from a higher level
of control, only to be stopped by a controller which hasn't been programmed
to deliver that control method.
Polytechforum.com is a website by engineers for engineers. It is not affiliated with any of manufacturers or vendors discussed here.
All logos and trade names are the property of their respective owners.