Predicting Pitting Potential of Additively Manufactured Stainless Steel using Machine Learning

The heterogeneous corrosion response of metal additive manufacturing (AM) parts caused by the variability in the printed parts hinders their broad adoption and implementation. Existing corrosion response characterization protocols rely on experimental observations that, while useful, are limited to providing qualitative guidance on the performance of new printed parts. In this work, a protocol for establishing a robust and predictive model that links the corrosion behavior to its corresponding processing parameters and as-printed part descriptors is developed. The developed protocol distills a set of features from the AM inputs with unsuper-vised learning and subsequently uses ensembling models to build a robust predictive model for the corrosion behavior of AM parts. This protocol is validated by predicting the electrochemical breakdown potential of as printed, AM stainless steel 316L as a function of AM printers and heat treatments. The developed framework showcases a practical pathway to leverage prior experimental data to rapidly estimate corrosion response in AM stainless steel.