Rapid Data Acquisition and Machine Learning-Assisted Composition Design of Functionally Graded Alloys Via Wire-Feed Additive Manufacturing

The application of artificial intelligence in materials discovery and design is hindered by the lack of high-quality datasets, which necessitates the development of rapid techniques for acquiring reliable data. Wire-feed additive manufacturing (WFAM) offers a promising approach for fabricating functionally graded alloys with precise composition control, thereby enabling generating extensive datasets to investigate process-structure-property (PSP) relationships and facilitating machine learning-assisted alloy design. Leveraging high-throughput experiments, calculations, and genetic algorithms applied to WFAM-built graded alloys, a machine learning (ML) model was developed based on a database with 32 material descriptors and hundreds of data entries, capable of predicting hardness and porosity. The ML model has demonstrated its efficacy by successfully designing a gradient alloy with enhanced properties, and thus can be used in the functionally graded alloy printing from P91steel to 740H superalloy. This work represents a significant advancement in the application of coupling machine learning and additive manufacturing for materials design. In addition, it offers a rapid and efficient approach for exploring process-structure-property relationships and accelerating alloy development. However, a notable level of uncertainty has been observed in the tensile properties of the designed composition that blends 90 wt.% P91 steel and 10 wt.% 740H alloy. This phenomenon can potentially be ascribed to the larger size of the designed alloy build compared to the gradient print utilized for constructing the machine learning model. This necessitates the inclusion of uncertainty quantification and processing optimization techniques into this design approach. This research brings attention to the importance of considering the part size and thermal history on process-structure-property predictions, particularly in the scale-up design for WFAM.