The main goal of this self-organisation feature is to make the controller robust as much as possible to any changes. A trade-off exists between performance and robustness, since robust control designs are usually achieved at the expense of the resulting closed loop system performance (relative to a control design based on a perfect model). Advanced robust control system design methods have been developed to minimise this inherent performance/ robustness trade-off. Although robust design methods are currently limited to linear problems, nonlinear problems with model uncertainty can sometimes be approached by gain scheduling, a representative
set of robust point designs over the full operating envelope of the plant, thus decreasing the amount of model uncertainty that each linear point design must accommodate. Nevertheless, the performance resulting from any fixed control design is always limited by the availability and accuracy of a priori design information.
In this case, it is suggested to employ a LMI (linear matrix inequality) approach. However, if you are not familiar with this technique, it would be a little difficult.
see this:
Lin, C., Wang, G., Lee, T.H., He, Y. LMI Approach to Analysis and Control of Takagi-Sugeno Fuzzy Systems with Time Delay, Springer, 2007.
Donald Hubbard might help: "How to measure anything" It's not scientific, but it points to a lot of scientific findings. It does not tell you, how to do it but at least how to avoid pitfalls.
But the main question is: can we design a good fuzzy controller at the first stage?
In some cases (usually with simple problems), the controller’s design solves the problems, demonstrates a satisfactory performance, and does not need any further reconsideration. The goal of the second stage is to enable a wider class of problems to be solved by reducing the prior uncertainty to the point where satisfactory solutions can be obtained on-line but how???
After many manual tests, you can find the parameters of the fuzzy controller that guarantee the stability of the closed loop system. Of course, these parameters are not the best ones (only if you are so lucky). The first stage may help you in the next stage. Generally, the next one corresponds to an optimization of the controller parameters by the mean of heuristic or metaheuristic algorithms in order to achieve some requirements.