Multivariable digital and analog control strategies


Thread Starter

Mauricio Valdes

I want to know the concepts about analog and digital control and wich are the normally used strategies to improve a digital or analog multivariable control.


Steve Monnet


You have many strategies for multivariable control :

Fuzzy logic : use fuzzy algorithm and rules to evaluate the impact of each input in order to define the ouptut. This technology, used in
control since ~1980 is very powerful. You can modelize the expert knowhow into the controller (in fact the fuzzy logic is apparented to expert system in the sense that it is also based on rules
evaluation). For example some PID autotuning system are based on fuzzy logic. Fuzzy is also used for temperature controller. Many PLC
suppliers have developed fuzzy controller you can implement into a PLC (Siemens, Omron,...)

Expert system : Despite the fact that Expert system are often too slow to implement directly into a PLC, It is possible to configure it for higher level control (giving the preset for PID controller). Lilly has made some experiences related into speciallized press.

Control based on simulation : like the expert system this strategy is more often used as a top level control layer. It needs a strong effort initially to developp a simulation package for the process you want to monitor

Auto-adaptive PID : This startegy use some input from the process to parameter for normal PID loop (cascaded or not). In other term, you move the range of stability of a PID controller in function of other process variable. In general this strategy is useful for fast response time but it is very "tricky" to tune.

In general none of the above strategy is the "one way". You must choose the most adapted to your process.

If you want more information regarding this subject, feel free to contact me directly.

Steve Monnet
[email protected]
Thank you Steve. Your tutorial is wonderful in short form.
Mauricio, read me in all aspects of control I replied, you will find so many hints.
Read particularly few topics above 'Fast reaction'.
Steve is so right saying it takes lots of efforts and maths to develop a model. One would be wise considering decoupling of the dynamic. It's also good strategy. Don't embrasse more than what you can embark.
In the analog age, we would decouple as much as possible.
An example of multivariate process is the boiler drum level. If for instance, you would start a high pressure boiler operating near the critical point and start it from cold and using a single XTR (single calibration), then you would have to take into acount:
1 . the volume of entire contaiment (pipes, mud drum...) and the ratio to the drum level.
2 . the change of weight density of water vs temperature (at critical point she is les than one third than at cold, and in that vincinity she drops nearly vertical). Surely, one will spend time as there is need to approximate several functions (at least one of them is not approximable via math software available: the appropriate formula is needed for water ).
After, at operating conditions the steam demand (based on pressure) is feedforwarded into the mass flow rate of the feed water...and so on.

Whatever strategy, fuzzy logic or approximating feedforward algorithms it is advisable you leave it transparent to the reader, so that one can follow and check step by step (even for yourself later on).
Too much concatenation of variables may result in cacophony.
Like Steve says, if you do multivariate optimization, do it in separate level, completly indepentant of the manageable loop level.
Watch carefully the cracks in maths near you and review the numerical maths in your optimizing package.
I have seen thousands of man/hour of math cracks reinventing the wheel of a simple system (not so simple in fact). We never go back too low in our academics (background). Sometimes it is necessary to start again from scratch, and brain storm.

Best, better are relative words.
[email protected]