SBIR-STTR Award

Most existing simulation tools are deficient in modeling gas flow effects, which have been proven to significantly affect the build quality of metal additive manufacturing (AM). In Phase I, the research team will extend our current Integrated Computational Materials Engineering (ICME) modeling toolkit by including a multiscale and multiphysics simulation modules to predict the effect of gas flow on metal additive manufacturing (AM) processes for improving the quality of the parts. The framework carefully integrates the discrete element method, chamber-scale aerodynamics, and powder-scale thermal multiphase flow models to comprehensively resolve the multi-physics in the metal AM process and quantify the gas flow effects on melt pool dimensions, surface roughness, solidification rate, powder spattering, and pore formation/propagation. The developed framework will be thoroughly first validated on laser powder bed fusion (LPBF) using existing data and new experiments conducted by the team. Then, the model will be extended to model multi-layer multi-track LPBF and directed energy deposition (DED) processes. Physics-informed machine learning-based surrogate models will also be constructed to enable the performance driven process design. The developed framework will directly quantify the gas flow effects by taking AM parameters and material properties as inputs and help establish practical mitigation strategies for the gas-induced powder spattering and pore formation.

Benefit:
The uncontrollable internal defects and surface quality inherent to the current AM process hinder adoption of AM to critical applications. The research will result in a versatile, user-friendly, and computationally efficient toolkit for multiphysics simulations of metal additive manufacturing (AM) to optimize process parameters to enhance the fatigue performance of aircraft components. The research will advance the state-of-the-art numerical methods for modeling gas flow effects, and the resulting multiscale and multi-physics models will fill the technology gap of current simulation tools and dramatically enhance the predictive capabilities of metal AM processes with the local gas flow induced spattering. The high-fidelity models will deepen the understanding of complicated multiphysics phenomena in metal AM processes to capture the effects of gas flow on melt pool dimension, surface morphology, temperature profile, solidification rate, powder spattering, and pore formation. The integrated framework will also allow additive manufacturers to determine optimal gas flow parameters to minimize gas-induced defects while significantly reducing trial-and-error testing on expensive physical prototyping. The developed toolkit can be integrated with other physics models, such as microstructure models, to establish the in-demand process-structure-property relationship in metal AM. A machine learning-based reduced order model cast from the physics-based model and collected data can quickly evaluate the property uncertainty and gauge the fabrication parameters to reduce the level of defects leading to low development costs, faster production, and improved performance of metal AM parts.

Keywords:
gas flow, gas flow, laser powder bed fusion, powder motion, Direct Energy Deposition, Multi-Scale, Metal Additive Manufacturing, spattering, multi-phase modeling

Global Engineering and Materials, Inc. and Professor Jinhui Yan at the University of Illinois at Urbana-Champaign propose to develop an Integrated Computational Materials Engineering simulation toolkit for optimal tailoring of gas flow in laser powder bed fusion (L-PBF) and direct energy deposition (DED) to reduce defects (e.g., porosity and spatter) and surface roughness improve quality (e.g., microhardness and heat-affected zone). The proposed tool, Additive Manufacturing Gas Flow Simulator (AM-GFS), quantifies the gas flow characteristics such as nozzle flow in DED and gas circulation in PBF, and predicts defect/quality index for the component-level print as typical in the aircraft. The model capability highlights are summarized as follows: 1) multi-scale model that couples the gas flow phenomena in powder-scale and chamber-/nozzle-scale. 2) High-fidelity powder-scale physics modeling that resolves the laser absorption, molten pool, vapor jet, gas entrainment, and spattered particles. 3) Full validation using in-situ and ex-situ data (e.g., surface profile, spatter count, and molten pool size). 4) Physics-informed machine learning (ML) based surrogate models that are trained based on simulation data to fast produce process-to-defect relationship. 5) Cross-process models which are robust to accommodate both DED and PBF processes. The results from the AM-GFS tool will establish a process map that delineates the boundaries of high defect index region in a gas-flow parameter space. Such capability will accelerate the process design iterations to identify the optimal gas flow that minimizes defects.

Benefit:
The research will result in a versatile, user-friendly, and computationally efficient toolkit for virtual simulation of gas flow and defect formation in PBF and DED processes. Such capabilities will allow the AM process engineer to down select the parameter space to achieve optimal print quality, given a new print geometry, material, roster pattern, etc. The physics-informed machine learning model will efficiently produce indications of poor-quality region, and the process engineer can run simulation before the printing to reduce trial-and-error cost. The developed toolkit can be linked with other physics models to further predict the microstructure, mechanical properties, and fatigue performance, among other things. This toolkit can be applied to both additive component printing and cladding repair scenario. The virtual experiments will enable faster design and process validation and verification, leading to lowered development costs and adding the values to a long-term production cycle.

Keywords:
Surface Roughness, Direct Energy Deposition, gas flow optimization, laser powder bed fusion, physics-informed machine learning, defect