Modeling and prediction of surface roughness in belt polishing based on artificial neural network

Abstract

Surface roughness is a variable often used to describe the quality of ground surfaces as well as to evaluate the competitiveness of the overall polishing system, which makes it an ever-increasing concern in industries and academia nowadays. In this article, from microscopic point of view, based on the statistics analysis, and by the use of the elastic contact theory and the plastic contact theory, the model of the maximum cutting depth of abrasive grains is developed. Then based on back-propagation neural network, taking the maximum cutting depth of abrasive grains, the rotation speed of belt and the feed rate of workpiece as the input parameters, a prediction model of surface roughness in belt polishing is presented. The prediction model fully takes the characteristics of polishing tool and workpiece into consideration which makes the model more comprehensive. Compared with the model that takes the polishing force as the input parameter, the model in this article needs fewer experiment samples which will save the experiment cost and time. Moreover, it has a wider range of uses and is suitable for different polishing situations such as different workpieces and polishing tools. The results indicate a good agreement between the predicted values and experimental values which verify the model.