Unraveling process-microstructure-property correlations in powder-bed fusion additive manufacturing through information-rich surface features with deep learning

Abstract

A machine learning (ML)–based framework has been developed to optimize the process parameters and unravel the paramount process–microstructure–property (PMP) relationships rapidly and precisely, which is demonstrated using electron beam melting (EBM®)-processed Ti–6Al–4V alloy. The process maps are constructed using ML to discover the optimal process window based on the three criteria: fully dense (≥ 99.5% in relative build density), flat fusion surface and highest possible strength, without which the optimization outcome will be deficient. By using the ML-optimized process parameters, Ti–6Al–4V parts were fabricated with ultimate tensile strength ∼5% higher and ∼80% faster than the classical parameter setting developed by the original machine manufacturer. The α/β dual-phase microstructure and tensile deformation mechanism are further studied to understand the superior mechanical properties obtained in the ML-optimized sample. In addition, the information-rich top fusion surface is found to be an alternative route to implicitly materialize the microstructure-to-property relationships with a prediction accuracy of ∼98%. The full-scope surface and microstructure maps are then generated from the process parameters. Surface features that reflect superior and poor mechanical properties are particularly identified. Finally, the implications of the ML-based PMP framework with surface feature correlation are discussed in the context of improving resources utilization efficiency.