Neural Network Prediction of Surface Roughness with Bearing Clearance Effect

Abstract

In manufacturing industry, the quality of manufactured machine components, is determined by how well they follow a defined product's criteria for dimensional accuracy, tool wear, and surface finish quality. For this reason, manufacturers must be able to regulate machining processes to ensure improved performance and service life of engineering components. This research work presents a study on the optimization of machining parameters for mild steel using artificial neural networks (ANNs). The focus is on developing an effective and efficient machining technique for mild steel by leveraging the capabilities of ANNs to predict optimal machining parameters. To bridge the gap between laboratory figures, model-simulated values, and real-world application, experiments were conducted to obtain data used in the research analysis. Levenberg-Marquardt method were utilized to train the ANNs, with input factors like depth of cut, bearing clearance, cutting speed, and feed rate considered, while the surface roughness of the cut, normalized within 0 to 1 range. A statistical measure of the surface roughness predicted using indicated MAPE value of 0.002% while the correlation coefficient (R) was 0.99995. The results showed that ANNs are a viable machining parameter optimization method and can improve product quality, while providing significant economic and production benefits.