Abstract
In this study, a new technique of in-process evaluation of the grinding wheel surface is proposed. Some specified wheel surfaces are prepared as the references via the appropriate truing/dressing procedure, and grinding sounds generated by these wheels are discriminated by analyzing the dynamic frequency spectrum of 6-10kHz with a neural network technique. In the case of conventional vitrified-bonded alumina wheel, grinding sound can be identified under the optimum network configuration in such that learning rate is 0.0029 and number of hidden layer is 420. The resinoid-bonded CBN wheel is also discriminable with the grinding sound in higher frequency range. This system can recognize instantaneously the difference of the wheel surface in a good degree of accuracy insofar as the wheel conditions are relatively widely changed. In addition, the network can perceive the unlearned wheel condition as the nearest one.
