Grinding process control through monitoring and machine learning

The plunge grinding process has been investigated via the power spectrum of vibration signals. In order to predict the process evolution, certain pre-defined spectral attributes were extracted and process performance classes were assessed by an expert. Training examples, described in terms of the attribute values and corresponding classes, were submitted to an inductive machine learning system. As a result, classification rules were synthesized, predicting the grinding wheel performance from the spectral attributes. After refining training data, the classification accuracy of the induced rules was increased and their complexity reduced. The investigation provided a new insight into the problem domain by discovering attribute interrelations and their significance. Moreover, the obtained results are suitable for application in grinding process control.