An intelligent method to monitor the abrasive belt condition based on sound signals

Tool condition monitoring is essential for increasing performance of robotic grinding processes. A variety of methods have been explored to address this issue, but have limited success. This paper introduces an innovative method to monitor the abrasive belt condition quantitatively by using grinding sound signals. Fast Fourier Transform (FFT) and Discrete Wavelet Decomposition (DWD) are deployed to distinguish the belt-wear related signals. Sound features are extracted from the separated signals. Using these features, a back propagation neural network is developed to predict the index measure of grinding ability factor which quantifies the belt wear condition and hence the Material Removal Rate (MRR). The prediction result shows that the relative errors under different grinding forces are all less than 4%, and the proposed prediction method is robust and effective.