Abstract
With the increasing amount of historical fatigue data for advanced manufacturing processes, such as additive manufacturing, it becomes increasingly feasible to use statistical and machine learning approaches to garner deeper insights into the contributions to fatigue performance in order to improve the design for fatigue failure or processing route parameters. Prior to model development, aggregated datasets, whether compiled through manual or automated processes, require extensive verification and profiling to eliminate systematic errors and identify insufficiently investigated parameter combinations. Without these steps, the veracity of any model, especially black-box models, is dubious. Once the structure and patterns of the dataset are established, proper implementation of random imputation can be used to expand the amount of usable data. This verified and augmented dataset can now be subjected to various statistical tools whose role in data exploration will be discussed, particularly regarding the role of distinguishing porosity- and microstructure-driven fatigue failure data.
