Characterization of the grinding process by acoustic emission

The properties of a ground surface can be estimated on-line during manufacturing based on the analysis of acoustic signals emitted by the grinding process. This possibility is demonstrated using an experimental system comprising an external grinding machine, a data acquisition unit and an artificial neural network. In the initial phase of system application, an empirical model of the grinding process is formed in the memory of the neural network by self-organized learning driven by empirical data consisting of the acoustic emission spectrum and a surface roughness correlation function. After learning, the system applies the model to estimate the correlation function of the surface profile from the input acoustic emission spectrum. For this purpose, non-parametric regression, based on the conditional average estimator, is utilized. Experiments were done on the grinding of hardened steel workpieces by a corundum wheel. During formation of the model, the surface profile and its correlation function were determined off-line, while in testing system performance the surface correlation function was estimated on-line from the acoustic emission spectrum. With respect to the estimation error, three characteristic periods of the process were observed corresponding to grinding with a newly dressed, slightly worn, and worn out wheel. The best estimation is obtained during grinding by a slightly worn wheel.