How does fuzzy logic eliminate the need of tuning?
I'm not an expert, nor even moderately knowledgable about fuzzy logic control. It's important that I make that clear up front. I guess you can say my knowledge about fuzzy logic control is "fuzzy" at best.
I'm much more familiar with classical and modern control theory and application and always try to attack problems from that point of view first, and usually succeed with that approach. The one place where I find myself using fuzzy logic control is when trying to control for maximization of a variable, rather than regulation or tracking of a variable. Maximization is not usually a straightforward control problem; however, a very simple fuzzy logic rule set can be used to maximize variables (multivariable maximization too!) and this tends to be very robust to system parameter changes. I can't really explain why, other than to say the rule set itself does not relate to system parameters, but mainly to system inputs and outputs. The output and the past history of the control inputs are monitored (and classified with membership funcitons), and then a rule set determines how to vary the modify the inputs for the desired outputs.
Generally, the modern and classical control approach requires having a model of the system and then matching control parameters (gains, and compensations). If we don't have a good model, then we might use PID and go through a tuning process, but in the end we've matched the controller to the system. Of course, there are ways to allow classical/modern control systems to be adaptive or self tuning.
My primary basis for making the suggesting and the claim that it is possible to design a fuzzy logic controller for temperature that is robust to a wide range of system parameter changes is that it seems that most modern temperature controllers have switched to this method. In the old days, we used to have a general set of PID gains for our temperature chambers, but if a significantly different system was to be tested with a complicated temperature profile, we sometimes had to retune to get the best response. These new controllers seem to have eliminated that need, at least according to what I've read when we purchased a new thermal chamber recently.
I guess a general statement that we can make (maybe) is that the need to do tuning when using PID control is replaced by the need to make a good rule set when doing fuzzy logic control. I think even with fuzzy control there is some gross tuning that needs to be done. For example, the update rate of the fuzzy control and the system dynamics need to be compared at a high level when planning the system design.