Process Development in Laser Powder Bed Fusion at the Melt Pool Scale

Laser powder bed fusion (L-PBF) is a complex manufacturing process characterized by a range of defect formation mechanisms that are a result of systemic process parameter selection and stochasticity in process interactions. Process development and process map approaches directly identify and control systemic defect formation mechanisms, however, can be sensitive too or miss less-apparent stochastic process phenomena, requiring further development. This research uses semi-analytical modeling, in-situ monitoring, machine learning, finite element modeling, and post-process characterization to investigate process development with respect to process phenomena that are not directly captured by defect process maps which cause melt pool scale defects. Melt pool scale defects, porosity or non-desirable irregularities, occur at time and size scales of the L-PBF melt pool and include: keyholing, lack of fusion, balling, spatter generation and inter-scan temperature accumulation. Two materials were studied in this work: AF962, a low-alloy steel with no established parameter sets and Ti-6Al-4V, a popular aerospace alloy with established parameter sets and extensive literature for L-PBF processing. 

In chapter 2 the influence of single-track conditions, model fitting and lack of fusion boundary criterion for process map development of AF9628 is investigated. Single-track dimensions are sensitive to powder entrainment and denudation. Measurements of single-tracks depend on sampling methods implemented. Eager-Tsai (E-T) model fitting requires optimization schemes to find optimal model inputs that enable accurate prediction. Bayesian optimization was observed to be as effective and less costly to fit E-T models, compared to a full-factorial search. Lack of fusion criterions should consider the variability in melt pool dimensions to make accurate density predictions of L-PBF components. The impact on process map development is shown to be highly sensitive to single-track dimension variability. 

In chapter 3 the control of near-surface porosity with contour selection and surface preparation to improve fatigue life is investigated. Process map development and modeling generally consider infill related phenomena only, where the interaction of the melt pool at the geometric contour is less understood. Furthermore, fatigue life of L-PBF components is low and highly scattered compared to conventional materials, where near-surface defects at geometric contours heavily influence fatigue life. Fatigue life improvement was attempted through surface preparation and contour remelting strategies for defect-free and defect-laden fatigue specimens. For as-printed surface conditions, fatigue life cannot be influenced by contour control due to high surface roughness. In treated surfaces, where as-printed surface roughness was nearly eliminated, contour remelting had a slight observable effect on fatigue performance. In post-post process characterization, contour remelting was shown to remedy defects in defect-laden fatigue specimens. 

In chapter 4 the control and mitigation of undesirable temperature accumulation between scan tracks is investigated. Temperature accumulation can lead to changes in material performance as it introduces inhomogeneous cooling rates. An E-T model was created to simulate the cool down for scan track terminations at an adiabatic boundary and process maps were created to proposed increased skywriting times to mitigate temperature accumulation. The proposed strategy was investigated through finite element modeling and in-situ experimental observations with a light sensitive photodiode and high-speed camera. The proposed strategy was observed to effectively suppress temperature accumulation. 

In chapter 5 spatter ejections are quantified to identify process parameters that mitigate generation. Spatter is a stochastic defect causing mechanism which is difficult to predict and model through simulation alone, requiring extensive in-situ measurements for characterization. A machine learning network is trained to identify and track spatter ejections in high-speed video observations of the L-PBF process for AF962 and Ti-6Al-4V. Spatter mitigation is correlated to moderate build rates, less scan tracks, and low laser powers within the conduction mode melting regime.