Roughness Assessment for Machined Surfaces in Turning Operation Using Neural Network

Abstract

Feed forward artificial neural network has been applied to predict the quality of turned surfaces for two types of coated carbide inserts. Four networks were proposed for each insert. The networks have been trained and tested using a former experimental data. The input data, represented by cutting parameter values, and output data, represented by surface roughness, were fed into the network model. Each network has three layers adopted for prediction. The first one is the input layer which involves cutting parameters: cutting speed, feed rate, and depth of cut; the second layer is hidden layer consisting of two hidden layers. The third layer of the network is the output layer which gives the surface roughness value. Levenberg - Marquardt algorithm is used in the back-propagation algorithm to train these networks. The best result was obtained for networks which have (12) neurons in the first hidden layer and (9) neurons in the second hidden layer. These networks had given R^2=0.9902 and mean square error = 0.0033 for the first insert, whereas, for the second insert, R^2=0.9892 and mean square error = 0.0023. These networks were used to predict the optimum cutting parameters which give minimum surface roughness.