The great debate, FL vs PID. Take a deep breath..and hold on.
I am fan of expert system, and peter is trying to convince me of the opposite.
BTW no hard feelings as both types have advantages/disadvantages.
No hard feelings. This isn't a religious issue as you will see. I am quite willing to modify the PID in any way I think works best. I think a good debate is educational.
I think a PID is restricted to only three processes (P,I,D)
Strictly speaking you are right but like I said above. There are modifications and ways to augment PIDs.
where expert can have more,
I object to the word expert system being applied.
People can be experts not software. At least not yet.
The use of FL doesn't give anybody any knowledge about the system.
it also has the capability of having > 1 inputs. and outputs.
Yes, but there is nothing to keep me from using 2 PIDs and tricks to achieve the same thing. Low select is a common example of how one can combine two inputs. A cascaded loop is an example of two or more PIDs each with an output and a input.
Honestly, I prefer neural nets, NN, over fuzzy logic. I will get to that later.
I always cringe when I hear someone wants to learn how to tune a PID.
ONE DOESN'T TUNE A PID. ONE TUNES A SYSTEM.
Therefore one should understand their plant.
People ask how to tune a PID and so many give advice based on their own experiences. The problem is that the person asking often doesn't know there are many types of systems and each is slightly different.
Starting from simple to more complex
type 0 ( non integrating, temperature, velocity ) single pole.
type 0 single time constant with dead time, FOPDT
type 0 , two real poles
type 0, two real poles plus dead time. SOPDT
type 0, two complex poles like a mass on a spring.
type 0, 1 real pole and 2 complex poles, not rare but rare to identify as such.
type 1 ( integrating, position, tank level ) and a single pole. Simple motor model and tanks
type 1 2 real poles. Motor models that model inertia and RL time constants.
type 1 2 imaginary poles. hydraulic cylinders and motors.
I have all these cases worked out symbolically. This gives me the ability to handle non-linear systems.
So where do you begin with FL if you don't know what kind of system you are trying to control? I can look at a system and know which of those models will apply. Knowing that I know how to approach controlling the system. With FL I must use a lot of trial and error or make a FL system look like a PID.
Things I do to augment PIDs.
1. Add feed forward. This is my most common trick.
2. The I gain must be in the forward path ( uses the error between SP-PV ) the P and D terms can be in either the forward path or only the feed back path ( -PV )
3. The gains in the forward path can be different from those in the feed back path. This is handy to move zeros or make a notch filter.
4. Often we add a second derivative gain. Adding a second derivative gain is important when controlling hydraulic systems. Without the second derivative gain the bandwidth of a hydraulic system is limited by the damping factor* natural_frequency. In other words the second derivative gain extends bandwidth.
5. I sometimes use cascaded loops.
6. Use Low-select. This can be tricky if the low select is bi-directional. In other words a low or high select at the same time.
7. I can add a smith predictor to compensate for dead time. I don't see how you can do this using FL.
8 I can use model based control where the PID is actually feedback from a model instead of the actual feed back. This is handy when the feedback isn't good enough.
9 I can change gains on-the-fly as a function of many factors.
10 I think you brought up the case of a car following a line. I mentioned how the chain rules should be applied so that the PID is updated as it covers distance as opposed to time. The speed relates the distance to time. Here is another example of two inputs being used. One for speed and one for detecting the line.
11 PIDs can be incremental so that changing gains on-the-fly do not result in a step in the control output.
12 PIDs execute in just a few instructions and are very efficient relative to FL.
An expert would know how to use all these tricks. People using FL are making rules up as the go along. Where is the expertise in that?
What is sad is that most people only think about PI or PID control and they are stuck using the PID that comes with their PLC.
It is never trial and error, (well it is same learning type as PID also trial and error.) And i think you must have knowledge on the proces to control it.
PID is trial and error for most people because they don't truly understand it.
