The first approach comes from an expert systems domain. The reason for that is pretty obvious: any fuzzy controller is actually an expert system applied to control problems. All theoretical and practical methods of knowledge acquisition developed in artificial

intelligence and other sciences are to be applied here. It should be noted that by using linguistic variables, fuzzy rules provide a natural framework for human thinking and knowledge formulation. Many experts find that fuzzy control rules provide a convenient way to express their domain knowledge. So cooperation with the experts will be easier for a knowledge engineer.

The main question is:

How do we find the rules practically? I mean how should this process be organised?

I think two methods can be proposed based on work with the documentation or the experts themselves.

The first one is based on redeveloping manuals, operation instructions and any other documents available into the set of the rules.

Another way includes an interrogation of experienced experts or operators using a carefully organised questionnaire. Of course, a good knowledge engineer always tries to apply the combination of these two approaches. Another method is pretty similar to the first one. Here the rules are formulated by observing how a skilled operator controls the object or the process. In this case the operator should be an expert. As was pointed by Sugeno, in order to automate a control process, one can express the operator’s control rules as fuzzy if–then rules employing linguistic  variables. In practice, such rules can be deduced from observation of the human controller’s actions.  But the main challenge to design controller is there another method based on for example, some measured data????

Always the preferred method is using theory of sampling and applying inference theory for defining fuzzy rules