Publication

Abstract : In this work, an experimental setup has been established consisting of a cylindrical grinding machine with piezo-electric sensor for capturing acoustic emission and its related hardware and software for signal processing. Acoustic signals are captured for the entire grinding cycle until the abrasive grains of the girding wheel become dull. Surface roughness produced by the process is recorded at fixed time intervals from the beginning to the end of the grinding cycle. Various features of the acoustic emission signatures such as root mean square, amplitude, ring-down count, average signal level are extracted from the time-domain are compared and correlated with the surface roughness generated by the grinding wheel on the work-piece. Good condition and dull condition of the grinding wheel is predicted using machine-learning techniques such as decision tree, artificial neural network, and support vector machine. Results indicate that there is a strong correlation exiting between the acoustic emission features and the surface roughness produced by the grinding process. Support vector machine trained with cubic kernel is appears to be predicting the grinding tool condition with greater accuracy comparing with decision tree algorithm and artificial neural network considered in this study.