Developing statistical tools to analyze historical stress-controlled fatigue data in additively manufactured austenitic stainless steel

Abstract

The development of robust fatigue property data sets for additively manufactured 316L austenitic stainless steels is hindered by the small sample sizes, inconsistent reporting of material properties and processing conditions, and differences in testing procedures for current sources of testing data. By applying standardization and categorization protocols to account for these shortcomings, historical data sets were aggregated into a single large data set and analyzed in greater detail, providing new levels of insight into the stress-controlled fatigue behavior of additively manufactured materials. Using this single aggregated data set, scatter was quantified using an eigenvalue analysis, and a multi-variable statistical model for predicting the mean fatigue life was developed. Interactions between variables and the sensitivity to different loading conditions were then identified, with the scatter in the data being attributed to differences in testing and post-processing condition such as surface condition and heat treatment, and the model was used to identify the role that each variable plays in determining fatigue life. An analysis of the fatigue responses across the aggregated data set using clustering algorithms allowed two distinct trends to be identified and attributed to the differing roles of microstructure and porosity on the resulting fatigue lives across different equivalent stress amplitude ranges.