1992 年 58 巻 7 号 p. 1239-1244
The extraction of essential features from adjustment behavior based upon the experience of skilled operators in response to a constantly changing grinding environment constitutes an important requirement with respect to the optimization of grinding conditions. Accordingly, this was attempted by means of a neural network system. The present system was developed with a view to practical utilization in production plants, which requires system characteristics robust with respect to various adverse operating environments. In the present study, a learning method was devised so as to remain effective even in the event that a portion of the learning data is lacking because of uncorrected grinding parameters or erroneous input, and the effectiveness of this method was demonstrated. Furthermore, a method was devised whereby heterogeneous data resulting from modification of specific condition parameters or erroneous input can be automatically detected by the neural network prior to learning, and the efficacy of this method was also demonstrated.