On the Robustness of a Systematic Methodology of Fuzzy-Logic Modeling

Links

Abstract

This thesis presents a detailed investigation of the robustness of fuuy-logic models. The key problem in fuzzy model generation is to find a suitable structure based upon a good model can be found. However, regardless of the modeIing approach, three as yet unsolved problems will always anse after developing the model: (i) no guarantee that the developed model has a satisfactory continuous behaviour: (ii) the model may show a "poor" performance if the data used to build the model is contaminated with noise; and (iii) the model may show a "poor" generalization capability if used to predict data that has not been used in the model generation. Each of the above problems is investigated in the context of the fuzzy-logic modeling approach.

To perform this investigation, a definition of the mode1 robustness is introduced and the systematic methodology of fuzzy-logic modeling (Emami et al., 1996a) is reviewed. The goal is to impr~ve this rnethodology with regard to the above-mentioned problems. First. the continuity behaviour of the fuzzy reasoning mechanism is investigated. As a result. sorne conditions on the reasoning parameters are introduced to adjust the sensitivity of the inference mechanism to account for any deviation in the inputs. Second, an irnproved noise rejection clustering dgorithni is introduced. The improved algorithm overcomes the pro blems of the tradi tional clustering algorithms and introduces a new criterion to cut off the noise in the data. Third. the generalization capabiIity of the hzzy logic model is investigated and some conditions on the input-output membership functions are introduced to improve the model behaviour when extended to a new set of data that has not been used in the model construction. AIso. two different evaluation criteria are snidied to validate the developed fuzzy-logic models.

Finally, the results are applied to two reaI applications. The first application is a well-known gas furnace plant. The second application is a 4 d.o.f robot manipulator developed in the Robotics and Automation Laboratory at the University of Toronto.

Cited Works (1)

Year	Entry
1974	Akaike H. A new look at the statistical model identification. IEEE Trans Automat Contr. December 1974;19(6):716-723.

Cited By (1)

Year	Entry
2000	Hasanaj E. Experimental Investigation of the Boundary Layer-Based Sliding Mode Control and Single-Input Model-Based Sliding Mode Fuzzy Logic Control With Applications to Robot Manipulators [Master's thesis]. University of Toronto; 2000.