| Abstract Scope |
Machine learning (ML) approaches have been used to develop and demonstrate a new, additively manufactured, blown powder directed energy deposition (DED-LB), functionally graded material (FGM) for hot-and-harsh gas path (HGP) jet engine components. A full-scale 737-like near-net-shape jet engine case made of Inconel 718 on the cold side and René 41 on the hot side was built. This approach was shown to provide static and dynamic material properties that are superior to a rotary friction welded baseline. Additionally, after production efficiencies are applied, this approach is projected to provide a 10% cost reduction relative to forged components. The machine learning approach combined iterative design of experiments and transfer learning to drastically accelerate the development of this process. First, process parameters for single-pass walls made of Inconel 718, a ‘weldable’ gamma’’ nickel alloy, were developed in 7 weeks without prior knowledge. Next, using the Inconel 718 data as prior knowledge in a transfer learning approach, single-pass wall parameters for René 80, an ‘unweldable’ gamma’ nickel alloy, were developed in only 5 weeks. Using the combined dataset, René 41, a moderately ‘weldable’ gamma’ nickel alloy, was developed in 3 weeks. This allowed the single-pass wall Inconel 718 to René 41 FGM parameters to be developed in 2 weeks. Finally, the FGM parameters were tuned to the thick wall parameters used in the demonstration build in only 2 weeks. This approach can be applied to a host of additional materials. |