The paper focuses on the problem of prediction of surface roughness in AWJ process and contribute to the online monitoring of the hydro-abrasive material disintegration process and its possible control. The main scope of the paper is to contribute to the usage of an artificial neural network as a decisive part in the surface roughness prediction and to outline a suitable online control mechanism. In paper a series of experiments were conducted to predict surface roughness and to use phenomena like acoustic emission and vibrations that accompany the cutting process to use in a possible process control. The model of artificial neural network was created in the MATLAB environment. In total, 150 configurations of multilayer perceptron with different configurations of numbers of neurons in hidden layers were developed. Two training functions, the Bayesian regularization and the Levenberg–Marquardt algorithm, were used during the network training. The results of the realized experiment have shown that the network with feedforward topology is able to predict correct value of the profile roughness parameter.
Design of the Model for the On-line Control of the AWJ Technology based on Neural Networks
RUGGIERO, Alessandro;
2016-01-01
Abstract
The paper focuses on the problem of prediction of surface roughness in AWJ process and contribute to the online monitoring of the hydro-abrasive material disintegration process and its possible control. The main scope of the paper is to contribute to the usage of an artificial neural network as a decisive part in the surface roughness prediction and to outline a suitable online control mechanism. In paper a series of experiments were conducted to predict surface roughness and to use phenomena like acoustic emission and vibrations that accompany the cutting process to use in a possible process control. The model of artificial neural network was created in the MATLAB environment. In total, 150 configurations of multilayer perceptron with different configurations of numbers of neurons in hidden layers were developed. Two training functions, the Bayesian regularization and the Levenberg–Marquardt algorithm, were used during the network training. The results of the realized experiment have shown that the network with feedforward topology is able to predict correct value of the profile roughness parameter.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.