Constructing process maps for pulsed wave laser additive manufacturing with interpretable machine learning

Abstract

Oxide dispersion strengthened (ODS) steels have found increasing applications in advanced fission and fusion reactors due to their excellent strength and resistance to radiation damage. The traditional powder metallurgy (PM) methods for fabricating ODS steels are costly and time-consuming. To address these limitations, attempts have been made to fabricate ODS steels with emerging additive manufacturing (AM) technology, such as laser powder bed fusion (L-PBF). However, up to now, the AM-fabricated ODS steels have not been able to match the quality of those from the conventional PM processes. In this work, we explore pulsed wave (PW) laser AM, as opposed to the conventional single mode continuous wave (CW) laser AM, for ODS steel fabrication. In particular, based on available experimental data of PW laser fabricated ODS steel parts, we construct a two-dimensional (2D) process map using two PW features (engineered using operating parameters) to define and visualize the parameter space that is directly linked to the quality assessment of the parts. Additional process maps are constructed to quantitatively model the density and nano-hardness of the parts. These process maps are constructed through rational feature engineering, supervised feature selection, and interpretable machine learning techniques. The constructed process maps for quality classification, density and nano-hardness regression analyses are validated by experiments. The process maps help us further understand the physical phenomena that produce good or bad quality parts.