Deep convolutional neural network-based in-process tool condition monitoring in abrasive belt grinding

Abstract

Abrasive belt grinding has attracted attention in recent years in both industry and academia due to the rapid development of abrasive belts. In-process tool condition monitoring in abrasive belt grinding is difficult due to the large and unknown number of abrasive grains with variable and stochastic cutting geometries especially when the monitoring utilizes complicated sound signal for functionality. To monitor the wear of an abrasive belt, a new method using the deep convolutional neural network(DCNN) is proposed to identify the wear state of an abrasive belt based on sound signals. To comprehensively evaluate the recognition result of the belt wear state, one-level accuracy and precision are proposed, and the accuracy, one-level accuracy and precision of the method proposed in this paper are 82.2%, 97.6%, and 0.863, respectively. Compared with traditional methods, the results of this study infer that this method based on the DCNN can automatically and simultaneously search for the features of grinding sounds that are sensitive to belt wear in two dimensions, the time-domain and frequency-domain. The above characteristics of the DCNN are very suitable for extracting the features of the nonstationary sound signals that are produced by the alternate cutting process of multiple abrasive grains.