Building of Grinding Models Employing Trainable Pattern Classifying Technique
Construction of Grinding Data Base System (1st Report)
1985 Volume 51 Issue 9 Pages 1725-1731
Details
1985 Volume 51 Issue 9 Pages 1725-1731
The present paper deals with the reliable mathematical model which is rendered indispensably necessary for the construction of the grinding data base system. The built-up mathematical models respond to a request of (1) the stable grinding operation data, (2) the grinding burn data, (3) the chatter data, (4) the surface roughness data, (5) the residual stock removal data, and (6) the out-of-roundness data, respectively. Each model is expressed by the discriminant function employed in the trainable pattern classifying technique and given by an equation of the form
W1 (log Dt) 2+ W2 (log Df) 2+W3 (log Dt) (log Df) +W4 (log Dt) + W5 (log Df) +W6 (log Pw) + W7 (log Vs) + W8 (log Rv) + W9 (log Rd) + W10
where Dt, Df, Pw, Vs, Rv, and Rd are dressing depth of cut, dressing speed, depth setting, wheel speed, speed ratio and diameter ratio, respectively, and all of them are the direct control variables by the operator on the machining field. The important features of the mathematical model are the reliability of the predictions and the simplicity in the computation. It is characteristic of our mathematical model that the classification of the grindability can be 'given by only evaluating the sign or the largest value of the discriminant function. And the minimum reliability of approximately 80% can be attained in the classifications, though too much grinding data are not required of us for adjusting the coefficients (W1, W2…, W10).
