Metal additive manufacturing is known for the fabrication of near-net shape parts, but it also allows the fabrication of materials with desired microstructure and properties. These benefits come at a cost: the challenging process control required to manufacture parts within given specifications. Therefore, a plethora of processing parameters must be identified. A strategy to minimize this cost is the application of machine learning to construct efficient predictive surrogate models that rely on a dataset consisting of a minimum number of data points. This approach suggests an interaction between the surrogate model and dataset, in which the model assesses the quality of the dataset and predicts which additional measurements can improve the dataset in the most efficient manner. This feedback is of particular importance when the additively manufactured parts must meet property specifications not only on average, but also within a given variance or uncertainty.

Here, we demonstrate a machine learning (ML) approach, which employs a heteroscedastic Gaussian process (HGP) model, from a dataset based on laser powder bed fusion of a glass-forming alloy at varying processing parameters. Using amorphicity as the microstructural descriptor, we train the model on our Zr52.5Cu17.9Ni14.6Al10Ti5 (at.%) alloy dataset. The HGP model not only accurately predicts the mean value of amorphicity, but also provides the respective uncertainty. The amorphicity shows a heteroscedastic dependence of the uncertainty distribution on the processing parameters, which the model captures. The identification of the aleatoric and epistemic uncertainty contributions allow to assess the intrinsic inaccuracies of the dataset. This HGP model approach provides a consistent way toward systematic improvement of ML-driven additive manufacturing processes.