Machine learning-driven optimization in powder manufacturing of Ni-Co based superalloy

Abstract

The process parameters in powder manufacturing must be optimized to produce high-quality powders with desired sizes depending on the use. Machine learning-driven optimization was applied to determine promising gas atomization process parameters for the manufacture of Ni-Co based superalloy powders for turbine-disk applications. Using a Bayesian optimization without expert assistance, starting from just three sets of data, three optimization cycles were used to determine the gas atomization process parameters. In particular, we determined the melt temperature and gas pressure that could achieve a 77.85% yield (size: <53 μm), compared to the 10–30% yield that is generally achieved. This substantial increase in yield enabled us to successfully reduce the manufacturing cost by ~72% compared with that of a commercial powder.

Keywords

• Powder metallurgy
• Bayesian optimization
• Gas atomization
• Ni-Co based superalloy
• Turbine disk