High Entropy Alloys Amenable for Laser Powder Bed Fusion: a Thermodynamics Guided Machine Learning Search
Ran Duan,Jintao Xu,Yakai Zhao,Qingjun Zhou,Zhenyu Yan,Yong Xie,Peng Dong,Lifang Xu,Kai Feng,Zhuguo Li,Xiubing Liang,Upadrasta Ramamurty
DOI: https://doi.org/10.1016/j.addma.2024.104195
IF: 11
2024-01-01
Additive Manufacturing
Abstract:Although there is considerable research interest in high-entropy alloys (HEAs), only a small fraction of the potential compositional space was explored hitherto. While additive manufacturing techniques such as the laser powder bed fusion (LPBF) can be gainfully employed to accelerate discovery of new HEAs with promising properties, the printability of such alloys could restrict such an exploration. Keeping this in view, a machine learning (ML) based alloy design strategy, which takes thermodynamic parameters such as the liquid temperature range (LR), liquid viscosity (LV), average expansion (AE) during the solidification of the alloy, and shear/bulk moduli ratio (G/B) of solidified alloy into account, is developed for assessing the printability of HEAs using LPBF. First, a printability evaluation chart that has six different printing capacity regions is obtained by considering LR, LV, AE, G, and B. Then, ML is utilized to predict the candidate chemical compositions within them that are amenable to LPBF. On this basis, an alloy with the composition Co7.5Cr7.5Nb7.5Ni10.0Ti67.5 (at.%) was identified, and its process window range and printability were validated through experiments. The range of thermodynamic parameters suitable for the manufacturability was determined as LV >5.5 mPa•s, LR >1600 ℃, AE <25%, and G/B <0.5. The effects of thermodynamic parameters on the printing process window was systematically investigated. Furthermore, Shapley Additive exPlanations (SHAP) algorithm was utilized to evaluate the influence of various elements on the thermodynamic parameters, which could, in turn, guide the chemical composition design so as to achieve HEAs with good printability. This work provides the thermodynamic parameter index for LPBF and offers opportunities to use ML for accelerating the design of HEAs that are amenable to LPBF.