Comparison Between Feed-Forward Back-Propagation and Radial Basis Functions Networks for Roughness Modeling in Face-Milling of Aluminum

The technology of the cutting process has evolved substantially in terms of materials, tools and machines; however it is a necessity to develop models for control and optimization of the cutting processes because nowadays industry relies mainly on empirical data and heuristic solutions provided by shop-floor experts. Due to the complex relationship between the variables of the cutting process, application of artificial intelligence approaches is wide feasible as modeling technique and functional for controller development. This work, presents a design of experiments, data analysis and model comparison for surface roughness prediction in face-milling of aluminum 6061-T6 considering tool spindle speed, feed rate, depth of cut and entry angle of the cutting insert as input variables. Measurements of average roughness are performed, then acquired data are preprocessed using the Principal Component Analysis (PCA) and Response Surface Methodologies (RSM). Once the proper variable arrangement is defined, a comparison between Feed Forward Back Propagation Neural Network (FFBPNN) and Radial Basis Function Neural Network (RBFNN) is performed based on the correlation coefficient between predicted and measured data. Results showed that Neural Network arrays have better prediction behavior than those based on RSM.