Some Possibilities of Using Doe in Setting Ann Parameters: an Application in Modeling of Abrasive Waterjet Cutting

Artificial neural networks (ANNs) have been successfully applied for solving a wide variety of problems. However, determining of ANN architectural and training parameter values still remains a difficult task. This paper is concerned with the usage of design of experiment (DOE) method in order to determine parameter settings of multilayer feedforward (MLFF) ANN trained with backpropagation (BP) algorithm with momentum for modeling purposes. In this paper, a case study of abrasive waterjet (AWJ) cutting was used to find ANN parameters settings in order to develop high performance prediction model. For evaluating the predictive performance of ANN models, the combined mean absolute percentage error (MAPEcomb) is used as performance criterion. The ANN model was used for prediction of traverse rate of the separation cut based on material thickness, water pressure, and abrasive rate as input cutting parameters. The selected 3-8-1 ANN model showed high prediction accuracy with correlation coefficient of 0.999. Finally, the developed ANN model is given in the form of mathematical equation. The results indicated that ANN model is able to learn the relationships between AWJ process parameters. Also, in searching for ANN model of high performance, DOE method can be efficiently used in setting ANN architectural and training parameters.