Neural-Network Predicting of Surface Finish or Cutting Parameters for Carbide and Diamond Turning Processes

Abstract

Turning experiments, measurements of surface roughness of the turned surfaces, and neural network computing were conducted to predict surface finish or cutting parameters for carbide and diamond turning processes. A systematic approach to obtain an optimal network was adopted with the consideration of the effects of the network architecture and activation functions on the prediction accuracy of neural networks. Analysis of variance was conducted to evaluate the significance of the effects of 3- and 4-layer networks with sigmoid and hyperbolic tangent activation functions. Trial networks were further generated for searching the optimal number of neurons in the hidden layers. The versatility of neural networks was further explored by retraining the networks by including a surface roughness parameter as one of the inputs and placing depth of cut and feed rate as the outputs, and the results were promising.