| Abstract Scope |
Effective process windows require rigorous statistical context. Process stability regions for the fatigue life of laser powder bed fusion (PBF-LB)-produced Ti-6Al-4V are identified via high statistical power, nested extreme value distribution parameter estimation. Specific focus is given on simultaneous variable screening and confounding analysis for numerous independent, physics-informed, and feature-engineered variables. Such variable breadth is enabled by explicit retention of data heritage/lineage from a simple, extensible, and robust data collection, alignment, and cleaning system. This system successfully structured data for a multi-year, multi-university NASA ULI program across multiple machines and participants, despite personnel churn and asynchronous data generation.
The resulting model yields expected life trends with respect to laser power and laser velocity, however, also includes simultaneous interactions in multiple other variables that would otherwise distort a two-dimensional view of the process window for any given stress level. Discussion on life prediction maps with practical implications for operations is included. |