Online Machining Optimization with Continuous Learning conditions (v 1 , f 1 , T 1 , F 1 ), (v 2 , f 2 , T 2 , F 2 ), and (v 3 , f 3 , T 3 , F 3 ) are selected to optimize the function F given by Equation (5). The source code is written in MATLAB 7.2 and Pentium-4 with RAM 512 is used. The computational time required is less than a second. 5. DESIGN OF VIRTUAL LATHE The effectiveness of the model is tested by con- structing a semi-virtual lathe. It consists of two modules: (i) Surface roughness prediction module based on neural network and (ii) Tool life predic- tion module based on empirical relation. For given process parameters, the machine provides center line average (CLA) value of surface roughness and tool life. 5.1. Neural Network RBF in terms of accuracy with lesser amount of training data. Hence, it is preferred to RBF network for prediction of surface roughness in this work. Even though researchers (Chryssolouris & Guillot, 1990; Feng & Wang, 2003) have different opin- ion about the performance of the neural network model, the effectiveness of the network depends upon various network parameters such as, train- ing and testing data, etc. Kohli and Dixit (2005) proposed a method for neural network modeling to predict surface roughness using comparatively lesser number of experimental data sets. In this work, the experimental data are selected from the work of Kohli and Dixit (2005) for dry turning of steel using carbide tools. The cutting speed, feed and depth of cut are considered as three independent variables on which the surface roughness depends. Therefore, three input neurons of a neural network are cutting speed (v), feed (f), and depth of cut (d). The network requires 21 data