A mel-frequency cepstral coefficient-based approach for surface roughness diagnosis in hard turning using acoustic signals and gaussian mixture models

During the last years, notable efforts have been made to develop reliable and industrially applicable machining monitoring systems based on different types of sensors, especially indirect methods that does not required to interrupt the machining process. As the main objective in machining processes is to produce a high-quality surface finish which, however, can be measured only at the end of the machining cycle, a more preferable method would be to monitor the quality during the cycle. Motivated by this premise, results of investigation on the relationship between audible sound emitted during process and the resulted surface finish are reported in this paper. Through experiments with AISI 52100 hardened steel, this work shows that such a correlation does exist between the surface roughness and the Mel-Frequency Cepstral Coefficients (MFCCs) and based on that correlation, a new quality monitoring method is proposed using Gaussian Mixture Models (GMM). Obtained results show that this method can identify three different levels of surface roughness with an average accuracy of 98.125%.