Two-Phase Reverse Neural Network Approach for Modeling a Complicate Manufacturing Process with Small Sample Size

We proposed a hybrid two-phase neural network approach for modeling a manufacturing process under lacks of observations, which is designed for determining cutting parameters in wire- EDM (Electrical Discharge Machining). The first-phase neural network, 1-K-M net, is designed for characterizing input-output relationship between machining thickness and the corresponding cutting parameters. The second-phase neural network, M-K-1 net, is also designed for identifying input- output relationship between cutting parameters and machining thickness, which is reverse mapping relationship to the first-phase net. The first-phase net has 1 to M mapping structure while the second phase-net has M to 1 mapping structure. Using 1 to M mapping, approximate cutting parameters are roughly estimated for a given machining thickness of material. All possible cutting conditions are generated from the first-phase net output. The second-phase net is employed for selecting the best condition using M to 1 mapping structure. Experimental results are given to verify that the proposed method could determine cutting parameters in wire-EDM efficiently