Composite descriptors for cast iron microstructures: a data-driven approach to understanding microstructure–property relationships

Ductile cast iron (DCI) microstructure is inherently complex, consisting of numerous interrelated features such as graphite size, distance, shape, ferrite and pearlite content, as well as grain size. These parameters are conventionally determined and interpreted individually, which does not allow understanding their combined or collective impact on mechanical behavior. This paper tries to address this issue with an innovative data-driven methodology to characterize microstructure and mechanical response by capturing all interrelated individual parameters in single composite descriptors via Principal Component Analysis (PCA). Extensive experiments are conducted to fuel the data models. PCA transforms the intercorrelated microstructure features in new, uncorrelated composite microstructure descriptors. This enables better comparability between DCI microstructures by assigning a unique microstructure signature in PC1-PC2-space, which are interpreted physically as coarseness (PC1) and graphite content (PC2). A subsequent PCA including test conditions (temperature and strain rate) and correlation with mechanical properties showed that now PC1 and PC2 can be interpreted as a general composite microstructure and testing conditions parameter, respectively. The PCs show clear correlations with mechanical properties, especially ductility. Finally, a regression-based feature importance was used in order to quantify the influence of individual microstructure and testing condition parameters on the mechanical behavior. The yield strength showed a strong dependence on testing conditions, while tensile strength and elongation at fracture are dominated by microstructure variation. In general, graphite nodularity and pearlite fraction were identified as dominant features, while graphite diameter plays a subordinate role and ferrite grain size is negligible.