Abstract
In order to optimize accuracy and efficiency the grinding process, the grinding results of every possible combination of grinding conditions should be predicted and evaluated prior to actual grinding. Many kinds of grinding models and databases have been proposed for this, with one of the simplest and most popular being the database which uses the classical regression model. To enhance the flexibility of this type of database it is important that the data which is utilized for prediction and evaluation be that which has been collected through field testing. However there are many cases in which mne data is not rich enough for estimating the regression coefficients. The learning and prediction method which this paper proposes is applicable in cases where the amount of learning data is not sufficient for re-estimating the parameters of the regression model. In addition to the classical regression model this method utilizes the fuzzy linear model. The performance of learning and prediction algorithms are evaluated by computer simulation and case study.
