Development of Physics-Based Numerical Models for Uncertainty Quantification of Selective Laser Melting Processes

Increasingly demanding space missions require structures and equipment that perform better, are lighter, and more affordable than those produced using more traditional manufacturing methods. Additive manufacturing techniques, and Laser Powder-Bed Fusion (e.g. Selective Laser Melting, SLM) in particular, have a demonstrated capacity to manufacture such components. In fact, a recent report of the National Research Council1 emphasizes the potential of additive manufacturing for the development of aerospace systems with increased performance. However, the adoption of these relatively new manufacturing techniques for the development of space systems will require the integration of computational techniques in the certification process of such components. The primary goal of the research described in this report was to enable the characterization of the influence of process parameter variability on components manufactured with the SLM technique for space flight systems and their performance. To this end, a physics-based surrogate model was developed that enables the fast prediction of thermal field and melt pool geometry during the SLM process. Two variants of the model were constructed one for uniform properties the other for non-uniform properties. The development effort was informed by a joint experimental effort and by high-fidelity numerical simulations of the process. An Uncertainty Quantification (UQ) strategy was developed in which this physics-based surrogate model is calibrated with experimental data (e.g. track width) using Bayesian analysis. The low-order uniform property surrogate model is fast enough to enable extensive UQ analysis for a wide range of process parameters. It was found that the non-uniform property surrogate model is subject to conflicting constraints that affect its usability in the context of UQ. A case study demonstrating how that UQ strategy can be used to explore the prediction and uncertainty of surface porosity in two dimensions was conducted. The architecture of the suite of algorithms constituting the UQ framework, including the surrogate models, was rendered more user-friendly under the sponsorship of the Jet Propulsion Laboratory (JPL) and is now accessible to both NASA Ames Research Center (ARC) and JPL. The uniform property surrogate model could potentially constitute the basis for the development of forward-process control strategies. However, its validation is still limited, and a more extensive effort is needed in that area. The availability of the model permits a wide-ranging sensitivity analysis that is on-going under JPL sponsorship. The experimental investigation of the complex interactions between process parameters, powder characteristics, and dominating physical phenomena during the laser melting process showed that normalized enthalpy should be used as basis to judge the quality of single tracks. A design of experiment study conducted on three-dimensional builds showed that components printed with minimal energy inputs exhibit both higher top-surface longitudinal stress values and higher vertical displacement of free end after wire-EDM cut were performed on the samples. The work was conducted at UC Davis and UC Irvine in close collaboration with teams at the Lawrence Livermore National Laboratory and at NASA ARC. The collaborations with both LLNL and NASA ARC are continuing beyond the period of performance and these activities led to the development of new collaborations with JPL. This project constitutes a necessary step toward a scientifically based numerical prediction of the SLM process which will enable the certification of components manufactured using SLM techniques for use in mission-critical roles.