THE APPLICATION OF NEURAL NETWORK TO THE DEVELOPMENT OF SINGLE CRYSTAL SUPERALLOYS

The neural network (NN) method is applied to mechanical properties estimation and alloy development of single crystal superalloys. Databases have been constructed from previous publications and the Rolls-Royce materials database. The Bayesian neural network technique was used for the modeling of mechanical properties of single crystal superalloys in terms of alloy compositions and test conditions. Creep lives, yield strengths, and ultimate tensile strengths of various superalloys as a function of contents of alloying elements are estimated and analyzed. New alloys were designed by calculations of various properties as well as creep rupture lives for millions of compositions, followed by selection of optimized alloys. The developed alloys are made in single crystal form by directional solidification and tested. They exhibit excellent phase stability and creep rupture lives which are better than or equivalent to those of CMSX-4.