Session: 03-11-02: Laser-Based Advanced Manufacturing and Materials Processing II

Paper Number: 173105

Accelerated Smartscan: A Gpu Accelerated, Multi-Laser, Physics-Based Toolpath Optimization Algorithm for Achieving Tailored Properties in Laser Powder Bed Fusion 

Laser Powder Bed Fusion (L-PBF) is an additive manufacturing process that builds parts in a layerwise fashion by using lasers to selectively fuse 2D slices of the desired geometry into an incrementally-growing bed of metal powder.  L-PBF has attracted substantial industry attention in fields like aerospace and biomedical manufacturing because it can produce near-net-shape parts of unparalleled complexity with properties meeting or exceeding those of wrought components.  However, achieving the desired properties consistently during fabrication is an ongoing challenge; usually L-PBF parts must undergo extensive postprocessing to correct flaws like porosity, out-of-spec engineering properties, and excessive residual stresses.  Mid-fabrication L-PBF properties tailoring efforts are challenged on two fronts:  (1) L-PBF lasers cannot move fast enough to achieve arbitrarily-shaped heat fluxes that tailor properties and perform stress relief as done in Electron Beam PBF, necessitating complex toolpath optimization to achieve similar results.   (2) Length and time scale discrepancies exist between the microscale weld pool produced by the laser, and the overall part geometry, which exponentially increase the compute of this toolpath optimization.  The state of the art for mid-fabrication L-PBF property tailoring sidesteps the optimization problem by performing extra laser passes over each slice of the geometry after fusion, heuristically prescribing the powers at each extra pass, which succeeds in property tailoring and stress relief at the cost of adding excessive build time.  Our work leverages a two-laser machine architecture to parallelize these extra laser passes by using one laser to fuse material and the other to perform heat treatment in the fused material.  We call this algorithm SmartScan.  SmartScan designs the toolpath of the heat treatment laser, and the power at each movement, through rapid and accurate physics-based models of the process.  The model forecasts temperatures for all allowable combinations of movements and powers from the current laser position, and the algorithm selects the one that best meets the prescribed objective.  Here, the objective is provoking 10 microstructurally-relevant thermal cycles everywhere in the build, that is, forcing temperatures above a transformation temperature without melting.  Previous work showed SmartScan’s efficacy, however, curses of dimensionality in the physics-based model limited the size of parts we could test on.  Specifically, our prior modeling framework  was based around a finite difference method (FDM) description of transient heat conduction, and the storage and compute demands of an everywhere-regular grid of nodes with density sufficient to capture the melt pool, and uniform marching forward in time, limited us to millimeter-scale parts.  We were able to show statistically significant improvement in microstructural features, but could not test mechanical properties directly.  In this work, we change the modeled physics to an analytical description of the laser heat, which removes the curse of dimensionality by permitting an irregular grid and an irregular time step.  We also incorporate GPU acceleration to further increase the speed.  We use these speedups to generate SmartScan toolpaths for decimeter-scale test parts and assess the residual stresses and directionally-dependent tensile properties in the as-built state.  Residual stresses are measured by printing long cantilevers and measuring the deformation after they are cut from the plate.  Tensile properties are measured by printing standard tensile specimens oriented laterally and vertically to the build plate.  All samples are printed with SmartScan toolpaths, state of the art toolpaths (10 heuristically-designed follow up scans per layer), and a heuristically-designed single laser scan (the “nominal scan”). Relative to the state of the art and the nominal scan, we show that SmartScan produces statistically significant reductions in residual stresses and reduced tensile anisotropy while also reducing print time by 5x. 

Presenting Author: Nathaniel Wood Air Force Research Laboratory

Presenting Author Biography: Nathaniel Wood is currently a postdoc at the Air Force Research Laboratory through the National Research Council (NRC) postdoctoral research associateship program. His research interests are the quality control of metals additive manufacturing, control theory, numerical methods for solving partial differential equations, and heat transfer. He received both a B.S. and Ph.D. in mechanical engineering from Ohio State University, and a B.S. in applied mathematics from Ohio State University.

Authors:

Nathaniel Wood Air Force Research Laboratory
Nicholas Kirschbaum University of Michican
Edwin Schwalbach Air Force Research Laboratory
Sean Donegan Air Force Research Laboratory
Andrew Gillman Air Force Research Laboratory
Chinedum Okwudire University of Michigan