FINDING COMPOUND PROPERTY PATTERNS IN EMPIRICAL DATA SETS USING DISCOVERY METHODS. A. G. Jackson, Wright Laboratory, WPAFB, OH, USA, N. Kiselyova, Baikov Institute of Metallurgy, Moscow, Russia.
Extension of performance of materials beyond the current limits for bulk or thin films has become imperative because of demands for high performance. Frequently this means new materials must be designed that build on existing compounds in ways not easily determined using first principles computational approaches. Methods for exploring data bases to identify patterns of behavior are useful because of the classification and prediction possibilities they offer. The pyramidal net method has been applied to the prediction of new compounds based on known data on the elements of the compounds. Compound types A3BC15, ABF5, ABF6, ABO3 have been considered using several forms of feature sets that include a number of properties of the elements and of archtype compounds where known. A Table of predicted compounds resulted from the analysis.