Control system state-of-the-art

You've opened a pretty good can of worms. The answer depends upon what level of automation you wish to work with. The answer has varied over the past half century. There may now be cases, especially in chaotic controls, in which the question must be reversed to ask what humans can't do but automation does well.

Automated control is fairly easy if there is a reasonably simple and known two way relationship between one or several input variables and one or more controlled variables. It gets a bit more difficult when the relationship is not known, or tends to vary. If there is some level of unreliability in the relationship or variables, then it gets even worse. Automation does not necessarily handle implicit information well unless it is explicitly defined. It also does not acknowledge data it is not explicitly given. Automation does not work worth beans when the power goes out. Come to think of it, it sounds a lot like a 3 year old.

It is possible with enough engineering foresight to make a controller which is capable of at least low levels of "thought". The question often comes down to how much engineering time are you willing to spend and how well are you going to trust it. Another issue is the hardware constraints. Most control systems are purchased to just barely control a process (minimal effective design) and leave a little room for expansion. One could argue that the programming logic in many of them would not support anything close to artificial intelligence.

As time goes on, there is less and less that automation cannot do. CPU's are faster, memory is cheaper, user interfaces are easier (but this also can mean that the are much more limited), and the on board tools are more sophisticated. Still, the control systems don't often get together to trade ideas, exchange experiences, etc. They cannot go back years in memory to look for similar problems, and recall what the final solution was, especially if that was not carefully coded to them.

I-bots are now getting attention, but most control systems are not sophisticated enough to use them. Sometimes it doesn't get down to control theory itself, it really gets down to how sophisticated you wish to make a controller, and what level of tools are available. I've written programs which controlled entire plants with many units in them. Most of what I wrote could not be put into a control loop. My programs analyzed the plant, the operator requests, and the product requirements, then told the regulatory level controllers what to do. There were probably one or two traditional control loops in my programming, but a lot of various analysis routines.

Michael

Yet. Seen the latest from Mercedes?? :-)

Maintenance. That's what computers can't do..

Cameron:-)