Comparison Of Fuzzy Logic And Neural Network Models For Surface Roughness Prediction İn Turning Of AISI 304 Stainless Steel

Abstract

This study presents a comparative analysis of fuzzy logic and neural network-based predictive models for estimating surface roughness (Ra) in the turning of AISI 304 stainless steel. Experimental trials were conducted using a two-axis CNC lathe, with cutting speed, feed rate, and depth of cut as input parameters. A full factorial design of experiments was used, and Ra measurements were obtained using a Mitutoyo SJ-210 surface roughness tester. A Mamdani-type fuzzy inference system and a feedforward neural network (3-10-1 structure, trained using the Levenberg-Marquardt algorithm in MATLAB) were developed. The performance of each model was evaluated in terms of RMSE and R^2. The neural network model achieved an RMSE of 0.155 µm and R^2 of 0.98, slightly outperforming the fuzzy logic system. Results indicate that both models provide reliable surface roughness predictions and can support intelligent manufacturing systems.