Application of Neural Networks to Modeling Cut Surface Quality for Plasma Arc Cutting
Jiayou WANG, Kazuomi KUSUMOTO, Kikuo NEZU
pp. 191-197
DOI:
10.2207/qjjws.18.191Abstract
This paper introduces the application of neural networks to modeling cut surface quality for plasma arc cutting process. The neuro-model of cut surface quality consists of three parallel neural networks, respectively, called the cut shape neuro-predictor, and dross attached level and cut surface roughness neuro-estimators. A modified BP learning algorithm was used to train the neural networks. Implementation of the neural networks in the modeling is discussed in detail. Prediction applications of the neuro-model are described for various cutting conditions.
Tested and estimated results show the effectiveness and acceptable estimation accuracy of the modeling approach proposed. The neuro-model developed is applicable to the base metal of mild steel and the cutting conditions described. By using additional training data at any time, fine tuning and enlarging applicable scope can be done for the neuro-model.