Machine learning enabled discovery of new L-PBF processing domains for Ti-6Al-4V

While it is well established that the microstructure and mechanical properties of Ti-6Al-4V alloys are sensitive to their thermal history, a relatively small portion of the processing space has been explored. Additive manufacturing methods such as Laser Powder Bed Fusion (L-PBF) offer extreme flexibility in processing conditions, but are still primarily limited through Edisonian experiments. This implies that there are new processing domains that could yield Ti-6Al-4V with favorable mechanical properties or improved build productivity. Using efficient X-ray Computed Tomography methods to quickly screen the density and subscale tension tests to identify mechanical properties of L-PBF Ti-6Al-4V, we leveraged Gaussian Process Regression (GPR) machine learning models and Bayesian Optimization sampling strategies to efficiently explore the L-PBF processing space to discover new processing domains. This study revealed (1) a substantially larger processing window to produce dense material (porosity < 0.1 %) by varying power, speed, and hatch spacing, (2) an ability to tune processing conditions to achieve yield strengths ranging from 980 to 1095 MPa and elongation from 4.6 % to 13.5 %, (3) a unique high density processing regime associated with high laser power, high laser velocity, and small hatch spacing, which was previously avoided due to concern of balling instabilities in single line trace experiments, and (4) new processing domains with improved mechanical properties and deposition rates. The new processing conditions resulted in extreme microstructural variation with differences in grain morphology, size, and texture. The degree of texture and grain refinement exhibited in the reconstructed beta-grains of some samples partially explained the large amount of ductility and variations in strength in the tested samples, although more sophisticated microstructural representations are required to fully explain the observed behavior. This new processing window and the associated material properties are valuable for informing the design of Ti-6Al-4V parts built by L-PBF. Furthermore, the machine learning methods employed here are not limited to this alloy or L-PBF method and can therefore serve as a template for other alloy systems processed by additional additive manufacturing techniques.