Design of expts. (DOE) and artificial neural network techniques (ANN) were used to study packing of fused alumina powders composed of three different sizes of particles. Two techniques have been used for prediction of powder compact porosities in mixts. of three different-sized fused alumina powders. The first is the mixt. design technique that produces a polynomial model of the powder-packing system. The second is the ANN technique that is extensively used to model complex systems in many fields. Three sizes (3, 30, and 350 .mu.m) of fused alumina powder were mixed and uniaxially compacted in the form of cylindrical pellets to measure their packing ability in the green state. Porosities of the cylindrical pellets were generated by prepg. powder mixts. in such proportions that were planned using mixt. design, which is a DOE method. A multilayer feed-forward back-propagation learning algorithm was used as an ANN tool to predict porosity, which was the response variable. Based on the training data, an ANN model of porosity as a function of constituent mix proportions was created with low av. error levels (2.7%). The results indicated that the mixt. model and the ANN model provided good predictions for powder packing.
DOE and ANN models for powder / S., Akkurt; Romagnoli, Marcello; M., Sutcu. - In: AMERICAN CERAMIC SOCIETY BULLETIN. - ISSN 0002-7812. - STAMPA. - 86(7):(2007), pp. 9101-9111.