Online monitoring of diamond grinding wheel wear based on linear discriminant analysis

Monitoring the state of grinding wheel is helpful for quality control and efficiency improvement of large-scale optics. Various process signals are always used for the evaluation of wheel’s state. Acoustic emission (AE) is the most sensitive one to the interaction of tool and material. However, AE signals are stochastic and board-band. The difficulty of feature extraction seriously impacts the performance of data-driven models applied in AE-based monitoring. In this paper, linear discriminant analysis (LDA) is employed on frequency spectra of AE signals and a LDA-based monitoring method is proposed to monitor online the deterioration of wheel’s state. LDA is one of the supervised clustering methods. Samples with different labels are projected into a low-dimensional feature space for maximizing the distinction between classes. During grinding process, AE samples are acquired on some nodes with a fix interval of material removal volume. Successively executing LDA on the growing sample dataset and the change of projection pattern can give a clear representation of the transformation of wheel’s state. Experimental results show that different wear stages and self-sharpening of grinding wheel can be real time recognized during grinding process.