Prediction and control of surface roughness in CNC lathe using artificial neural network

In this study, a neural network approach is presented for the prediction and control of surface roughness in a computer numerically controlled (CNC) lathe. Experiments have been performed on the CNC lathe to obtain the data used for the training and testing of a neural network. The parameters used in the experiment were reduced to three cutting parameters which consisted of depth of cutting, cutting speed, and feed rate. Each of the other parameters such as too] nose radius, tool overhang, approach angle, workpiece length, workpiece diameter and workpiece material was taken as constant. A feed forward multilayered neural network was developed and the network model was trained using the scaled conjugate gradient algorithm (SCGA), which is a type of back-propagation. The adaptive learning rate was used. Therefore, the learning rate was not selected before training and it was adjusted during training to minimize training time. The number of iterations was 8000 and no smoothing factor was used. R-a, R-z and R-max were modeled and were evaluated individually One hidden layer was used for all models while the numbers of neurons in the hidden layer of the R-a model were five and the numbers of neurons in the hidden layers of the R-z and R-max models were ten. The results of the neural network approach were compared with actual values. In addition, inasmuch as the control of surface roughness is proposed, a control algorithm was developed in the present investigation. The desired surface roughness was entered into the control system as a reference value and the controller determined the cutting parameters for these surface roughness values. A new surface roughness value was deter-mined by sending the cutting parameters to the observer (ANN block). The obtained surface roughness was fed back to the comparison unit and was compared with the reference value and the difference surface roughness was then sent to the controller. The iteration was continued until the difference was reduced to a certain value of surface roughness which could be permitted for machining accuracy. When the surface roughness reached the permitted value, these cutting parameters were sent to the CNC turning system as input values. In conclusion, both the surface roughness values corresponding to the cutting parameters and suitable cutting parameters for a certain surface roughness can be determined prior to a machining operation using the ANN and control algorithm. (C) 2008 Elsevier B.V. All rights reserved.