INSTRUMENTATION OF A ROLL COMPACTOR AND THE EVALUATION OF THE PARAMETER SETTINGS BY NEURAL NETWORKS

A Fitzpatrick L83 Chilsonator was instrumented in order to understand and to optimize the roll compaction process using drum-dried waxy maiz e starch, a plastic deforming material as a model compound. The interr elation of the four adjustable roll compactor parameter settings namel y the velocity of the rolls OCS), the speed of the horizontal (HS) and of the Vertical screw (VS), and the air pressure (P-air) influenced t he compact and the granule quality. The granule quality was defined by the friability and particle size distribution. As a second order poly nomial was not successful to model the behaviour of the friability in function of the four roll compactor parameters, a Multilayer Feed-Forw ard neural network (MLF) was used. It was shown that the MLF network m odels the friability more accurately than a second order polynomial. T he HS and the P-air mostly influenced granule quality and should be ke pt at a high level. The VS had no significant influence on compact qua lity. (C) 1997 Elsevier Science B.V.