Combining fuzzy logic and neural networks for decision support in an industrial environment

Fuzzy logic and neural networks are two attractive techniques that have recently received growing attention in decision support for real world nonlinear problems. However, they have limitations. The biggest problem in fuzzy logic is to determine its knowledge database whereas a major problem in neural networks is to explain its conclusions. Therefore, it is necessary to find a method which is able to alleviate some of these problems.;An approach which combines the fuzzy logic and neural networks is proposed in this study. It uses the neural networks training function to learn the system behaviour based on existing system data, then applies the neural networks recalling function to generate the fuzzy knowledge databases.;Fuzzy knowledge databases for the cutting parameters selection in milling operations developed by the NeuFuz method and the manual method are presented. A user interface is developed on a UNIX platform in order to integrate the FDSS (fuzzy decision support system) with a CAD/CAM (computer aided design & computer aided manufacturing) system.;Three tests of this NeuFuz method for decision support problems are carried out. The NeuFuz method is first tested in the manufacturing field to select the cutting parameters. The second test is also in the manufacturing field to predict the pre-travel error of a coordinate measuring machine. The application of this NeuFuz method in an industrial domain is then undertaken. An estimation of the poles life for Hydro-Quebec is done by using this NeuFuz method. The NeuFuz method is justified in comparison with the real results. The most important factors which have an influence on the accuracy are the complexity of the problem, the number of the examples, the coverage of the examples and the definition of the fuzzy premises sets.