Prediction and optimization of surface roughness in additive manufacturing with data-driven multiphysics models

Z. Gan, K.K. Jones, Y. Lu, L. Cheng, J. Lua, G. Wagner, W.K. Liu
Northwestern University,
United States

Keywords: surface roughness, data-driven multiphysics models, additive manufacturing


Surface roughness of additively manufactured part significantly affects mechanical properties, e.g., fatigue resistance. We developed data-driven multiphysics models that provide rapid prediction of surface roughness, and thus enable efficient optimization and minimization of as-built surface roughness. Because no single simulation method can simultaneously capture both the powder scale and the part or coupon scale phenomena in detail, we developed an approach based on a combination of high and low fidelity models. The high-fidelity model simulates surface roughness formation at the powder scale through a multi-phase thermal-CFD computation. This provides detailed mechanism of surface roughness formation and has been validated against new and previously published experimental surface roughness data. To enable larger scale simulation with complex part geometry, we simultaneously developed an effective medium thermal-CFD model that can simulate larger volumes, more layers, and more complex toolpaths. This model is calibrated using the high-fidelity model and experiments through a mechanistic machine learning approach and provides predictive capability of part scale surface roughness patterns with a larger range of materials, process parameters, and toolpaths.