Additive Manufacturing Modeling and Simulation: Microstructure, Mechanics, and Process: AM Modeling - Machine Learning and Artificial Intelligence
Sponsored by: TMS Computational Materials Science and Engineering Committee
Program Organizers: Jing Zhang, Purdue University in Indianapolis; Brandon McWilliams, US Army Research Laboratory; Li Ma, Johns Hopkins University Applied Physics Laboratory; Yeon-Gil Jung, Korea Institute of Ceramic Engineering & Technology

Tuesday 2:00 PM
October 19, 2021
Room: A113
Location: Greater Columbus Convention Center

Session Chair: Jing Zhang, Indiana University - Purdue University Indianpolis; Brandon McWilliams, CCDC Army Research Laboratory; Li Ma, Johns Hopkins University Applied Physics Laboratory; Yeon-Gil Jung, Changwon National University

2:00 PM
Deep Reinforcement Learning for Defect Mitigation in Laser Powder Bed Fusion : Odinakachukwu Ogoke¹; Amir Barati Farimani¹; ¹Carnegie Mellon University
    Powder-based additive manufacturing techniques enable the construction of structures that are difficult to manufacture using conventional methods. In Laser Powder Bed Fusion, components are built by selectively melting specific areas of the powder bed. However, challenges lie in the widespread adoption of these methods, due to the tendency for PBF-produced parts to develop defects and inferior physical properties in certain processing cases. Therefore, a control policy for dynamically altering process parameters to minimize the occurrence of defect causing phenomena is necessary. We present a Deep Reinforcement Learning (DRL) framework to optimize an closed-loop control strategy that avoids defect formation based on the character of the melt pool. The generated control policy alters the velocity of the laser during the melting process to improve melt pool consistency and reduce overheating. The control policy is trained on efficient simulations of the temperature behavior of the powder bed layer under various laser paths.

2:20 PM
Online Characterization of Melt Pool Dimensions Using Acoustic Monitoring and Deep Learning: Evan Diewald¹; Christian Gobert¹; ¹Carnegie Mellon University
    In powder bed additive manufacturing, many complex process outcomes, such as part density, mechanical strength, and surface finish can be strongly correlated to the dimensions of individual melt pools. For a given set of input parameters, we can measure these dimensions by conducting single bead experiments and imaging cross-sections. However, complex laser-material interactions, degrading machines, and heat buildup lead to significant variability in melt pool sizes when fabricating real components. Thus, in situ characterization is an avenue toward localized defect detection and uncertainty quantification. High-speed video cameras can capture these dimensions accurately, but prohibitive data requirements and a limited field-of-view present scalability challenges for such a system. In this preliminary work, we use high-speed videos as the “ground truth” dataset for a regression model that predicts melt pool dimensions from registered acoustic signals, which can be used to monitor the entire build. Both LSTM’s and CNN’s show promising early results.

2:40 PM
Machine Learning – Assisted Navigation in the Additive Manufacturing Design Space: Maher Alghalayini¹; Surya Kalidindi²; Chris Paredis¹; Fadi Abdeljawad¹; ¹Clemson University; ²Georgia Institute of Technology
    Additive manufacturing (AM) has been the subject of active research in recent years. In laser powder bed AM, a laser beam is scanned across metal powder. Recent reports have shown considerable variability in the observable properties of the printed parts mainly due to AM process parameters. This significant variability makes it crucial for engineers to consider variability while quantifying the resultant properties. Here, we propose a sequential sampling design to drive the search towards processing parameters that are expected to yield optimal mechanical properties. The novelty of the proposed approach lies in its use of machine learning to consider both variability and uncertainty to propose the next search site and the sample size to be tested. Experimentally obtained AM process-property data of 316L steel are used to showcase our approach. This study is expected to yield a methodology to efficiently navigate the AM process parameter space towards optimal materials properties.

3:00 PM
Process Consistency in Laser Powder Bed Fusion Observed Through Large Scale Single Bead Melt Pool Measurements: Christian Gobert¹; Evan Diewald¹; Jack Beuth¹; ¹Carnegie Mellon University
    In laser powder bed fusion processes, variation in melt pool geometry can lead to the inclusion of porosity, possibly undermining the integrity of printed structures. Even at constant process conditions, deviations in melt pool geometry due to the stochastic nature of the laser material interactions can occur. The value and variation in melt pool geometry was assessed using single beads printed at the center and edges of the build area, such that the influence of laser incidence angle could be studied. Measurements of melt pool widths were obtained using automated optical imaging of build surfaces. The prohibitive cost of cross-sectional microscopy limited the number of measurements for melt pool depth, however a correlation between width and depth was calculated. From the measurements of melt pool widths and depths, the melt pool can be modeled as a random variable and incorporated into process parameter selection strategies.

3:20 PM
Melt Pool Scale Modeling of Austenitic Stainless Steel Solidification Features in Laser Powder Bed Fusion: Joseph Aroh¹; P. Chris Pistorius¹; Anthony Rollett¹; ¹Carnegie Mellon University
    Laser Powder Bed Fusion (LPBF) builds components by using a rastering laser which produces microstructures comprised of numerous weld pools. Due to the rapid speed of the welds in LPBF, nonequilibrium effects such as metastable austenitic solidification and solute trapping may occur in austenitic stainless steels because of solidification kinetics. To better understand the influence of these effects on the as-built microstructure at the melt pool scale, a computational framework consisting of a numerical thermal model, CALPHAD, and a dendrite growth model was developed for several stainless steels. The model was compared to both in-situ synchrotron x-ray diffraction experiments and post-mortem characterization of cross-sectioned single tracks. This work aims to elucidate the location dependent microstructural evolution of sub-grain features at the melt pool level to inform future alloy properties, laser parameters, and scan strategies tailored specifically for the LPBF process.

3:40 PM
Gas Adsorption Analysis in 3D Printed Metal Organic Frameworks: Tejesh Dube¹; Hye-Yeong Park²; Yeon-Gil Jung²; Jing Zhang¹; ¹Indiana University – Purdue University Indianapolis; ²Changwon National University
    Metal-organic frameworks (MOFs) are compounds made up of metal ions or clusters coordinated to organic ligands. These frameworks have porous structures and thus serve as suitable candidates for gas adsorption, purification, and separation applications. This study focuses on combining MOFs with 3D printing in order to realize a component design for gas adsorption. MIL-101 (Cr), a chromium-based MOF was synthesized in-house and coated on 3D printed PETG substrate. This was further analyzed for gas adsorption behavior, through a combined experimental and modeling study. The computational model illustrates that MIL-101 (Cr) shows increased outputs in adsorption of nitrogen as pressure increases, similar to the trend observed in experiments. The model also shows promising results for carbon dioxide uptake at low pressures and hence the developed MIL-101 (Cr) based components could serve as viable candidates for gas adsorption applications.