Deciphering mechanical heterogeneity of additively manufactured martensitic steel using high throughput nanoindentation combined with machine learning

Abstract

Microstructures of additively manufactured (AM) materials are highly heterogeneous, periodic, and hierarchical at various length scales, leading to complex mechanical response. In this study, nanoindentation trend analysis combined with machine learning (ML) was used to unravel the hierarchical and heterogeneous microstructures and associated multiscale mechanical responses in a laser-directed energy deposited low-alloy martensitic steel. At length scale encompassing several melt pools, periodic hardness fluctuations were attributed to variation in strain accommodation and phase transformation during primary solidification and thermal reheat cycles between the melt pools. The observed trends were a net balance between dislocation accumulation due to volume contraction during liquid to austenite solidification, and relaxation due to subsequent solid-state martensitic transformation during cooling. The hardness trends within a single melt pool (250–300 µm) were ascribed to cooling rate variations which manifests in form of changes in primary dendritic arm spacing. The inter-dendritic regions enriched with Cr and Mo demonstrated higher elastic modulus and hardness. In microstructural scale, variation in pop-in behavior in the nanoindentation load-displacement (P-h) curves was correlated to local microstructural heterogeneity in the indenter-material interaction volume (dendritic segregation + martensite matrix) using various ML-based classification techniques: decision tree, support vector machine and neural network. It classified the indentations into three broad categories: first set showing only one pop-in lying within the segregation channel with 10,000 <P/h< 13,000 N/m, while the second set with P/h <10,000 N/m lied on the matrix and third set with P/h >13,000 N/m with multiple popins lied on the matrix-segrgeation interface. This novel multimodal structure-property correlative framework, integrating nanoindentation trend analysis with ML, provides a high-throughput approach to unravel the complex mechanical behavior of AM materials across multiple length scales, representing a key advancement in the field.