About this Abstract |
| Meeting |
2025 AWS Professional Program
|
| Symposium
|
2025 AWS Professional Program
|
| Presentation Title |
Machine Learning-Accelerated Discovery of Alloy Compositional Spaces and MicrostructureBehaviors for Additive Manufacturing of Components in Harsh Service Environments |
| Author(s) |
Giacomo Salvatore Melaragno, Boian Alexandrov |
| On-Site Speaker (Planned) |
Giacomo Salvatore Melaragno |
| Abstract Scope |
Alloy design with Calculation of Phase Diagram (CALPHAD) approaches have been widelyused in industry for accurate prediction of microstructural behaviors. For exploring complex andhigh dimensional alloy spaces, however, CALPHAD simulations are computationally inefficient.To accelerate the alloy development process, a machine learning (ML) framework has beendeveloped to generate sparse deep neural network (DNN) surrogate models that delivercompositional ranges for desired microstructures. In the current study, the ML framework isbeing deployed and validated on a Fe-35Cr-45Ni (35/45) alloy system used for pipe componentsin polymer-producing olefin furnaces. To improve service life in these components, afunctionally graded material is being designed to enhance 35/45 oxidation and carburizationresistance on the outer and inner diameter surfaces with Si and Al additions, respectively. Atrained sparse DNN was used to generate compositional ranges that seek to maximize Si and Alcontent in the FGM alloy without forming harmful phases like G-phase or NiAl. The frameworkwas validated by characterizing button melts of select compositions in the 35/45 alloy systemgenerated using the DNN surrogate model. |
| Proceedings Inclusion? |
Undecided |