GENETIC PROGRAMMING AND CAE NEURAL NETWORKS APPROACH FOR PREDICTION OF THE BENDING CAPABILITY OF ZnTiCu SHEETS

ZnTiCu (zinc-titan) sheets with approximately 0,1% Cu and 0,1% Ti content are very widely used in the construction industry for roof covering, gutters, drain pipes, facing linings, connections, window shelves, decorative elements on roofs, art products, etc. Data on the production technology of zinc-titan alloy sheet and on its forming properties are very scarce and unreliable. Therefore they must be checked for each individual technological step and the conditions under which the metal sheet is formed. Their forming properties are influenced by many parameters, e.g. chemical composition, technological parameters of rolling, etc. Due to large number of influential parameters the desired mechanical properties of the metal sheet (e.g. bending capability) are difficult to monitor and to keep within acceptable technological limits. Rolling mills usually collect data on an individual batch (e.g. alloy composition, conditions in which the sheet metal has been rolled etc.) but in most cases the general approach assuring achievement of the desired forming properties based on the influential parameters of metal sheet production cannot be traced. Often it is not known which parameters are of importance. In such cases linear regressions methods are not efficient since the abundance of input parameters and their mutual influences make the determination of an adequately precise model impossible 1 . In the present work two different approaches based on experimental data on ZnCuTi alloy composition and on technological parameters of hot and cold rolling have been used to predict the metal sheet bending capability. The first one is the GP which belongs to the class of the methods of evolutionary computation 2-7 , and the second one is the CAE neural network, which has been successfully applied for solving many engineering problems e.g. 8-16 .