Optimizing the fuzzy-nets training scheme using the Taguchi parameter design

Fuzzy nets have been proposed to combine the learning ability of neural networks and the reasoning ability of fuzzy logic to deal with complex control systems. This paper presents a systematic way of identifying the significant factors and optimising the performance of a fuzzy-nets application. To present the methodology, a model of a truck backing up has been evaluated. Four factors were considered: 1. The number of training sets. 2. The number of fuzzy regions. 3. The membership functions. 4. The fuzzy reasoning methods which would affect the performance of the fuzzy-nets training scheme in nonlinear applications. The Taguchi parameter design was implemented with an L-9 (3(4)) orthogonal array to identify the optimal combination for training consideration. Both raw and signal-to-noise (S/N) ratios were evaluated to identify, the optimal combination for the performance of fuzzy-nets training with very limited variation. The performance of the proposed fuzzy-nets scheme for the model of the truck backing lip was represented by the average errors between the truck and loading dock: 0.178 units and 0.204 degrees. The results demonstrate that the Taguchi parameter design is a robust approach for optimising the performance of the fuzzy-nets training scheme.