Prediction of Roughness Heights of Milled Surfaces for Product Quality Prediction and Tool Condition Monitoring

Abstract

The objective of this research is to predict the roughness heights of milled surfaces, which indicates product quality and tool conditions. Two experiments are carried out to evaluate relevant factors such as vibration, force, and surface roughness. The purpose of the first experiment is to find out the limits of the machining variables compared to the constraints of the materials. The purpose of the second experiment is to identify, collect, and compare how each factor affects product quality and tool conditions. Based on this study, the vibration, force, and surface roughness are good indicators for tool conditions. When the magnitudes of the vibration and force increase, the surface roughness also increases. The increase in surface roughness with constant cutting parameters indicates the degrading of product quality and the decrease of the tool life. Thus, the variables, such as vibration and forces, are used as the inputs, and the surface roughness is used as the output of neural networks. By optimizing the network variables, it has been found that a 4,4,8,1 neural network can achieve the least absolute error, and accurately predict the actual roughness heights collected from the experiment. The minimum error of the prediction of surface roughness is 0.11%, the average error is 2.11%, and the maximum error is 6.98%. The prediction of surface roughness of milled surfaces is very important for the product quality prediction and tool condition monitoring.