| Abstract Scope |
A large body of published fatigue data for additively manufactured Ti-6Al-4V exists but is not easily incorporated into existing statistical and machine learning tools, limiting the ability to develop robust design allowables. While both manual and automated data aggregation techniques can provide largely error-free stress amplitudes and cycles to failure, the inconsistent reporting of meta-data such as material properties, testing procedures, and post-processing minimizes the ability to extract larger trends and identify sources of uncertainty. In order to improve the quality of the aggregated meta-data, a data processing procedure which handles verification, profiling, and imputation of meta-data values was developed. Implementing this procedure revealed several systematic problems in previous aggregation efforts, which once addressed, yielded a more reliable dataset which could eventually be used to implement a database. Profiling the meta-data revealed important structures and patterns which place limitations on how the dataset could be used, and methods to use random imputation to expand the amount of usable data and meta-data during eventual modeling. |