Additive Manufacturing Benchmarks 2022 (AM-Bench 2022): Data-Driven Approaches I
Program Organizers: Brandon Lane, National Institute of Standards and Technology; Lyle Levine, National Institute of Standards and Technology
Thursday 10:30 AM
August 18, 2022
Room: Cabinet/Judiciary Suite
Location: Hyatt Regency Bethesda
Session Chair: Brandon Lane, National Institute of Standards and Technology
10:30 AM Invited
Benchmarking Machine Learning Modeling Approaches to Additive Manufacturing Research, Development, and Qualification: Aaron Stebner1; 1Georgia Institute of Technology
The prevalence of using machine learning models as research and technology innovation methods in the field of additive manufacturing is growing exponentially, as measured by submissions to the journal Additive Manufacturing. Collectively, researchers are incorporating such modeling approaches into nearly every aspect of AM spanning the process-structure-property design space, ranging from discovery to optimization to qualification problems and challenges for materials, machines, parts, systems, and more. However, the variability in documenting and benchmarking such approaches and their impacts is currently huge, resulting in great inconsistency and lack of comparability or adoptability of much of the work. In this presentation, the fundamental requirements of statistical modeling will be reviewed. Case studies and experiential learning will be used to motivate a proposed standardized framework for documenting machine learning methodologies. A list of priorities for creating and curating benchmark datasets that will enable greater technological advancement and understanding will be suggested.
11:00 AM
Physics-Based Compressive Sensing and Physics-constrained Dictionary Learning to Monitor Laser Powder Bed Fusion Process: Yanglong Lu1; Sungjin Hong2; Sung-Hoon Ahn3; Yan Wang4; 1Hong Kong University of Science & Technology; 2Georgia Institute of Technology; 3Seoul National University; 4Georgia Institute of Technology
The variability of qualities in additively manufactured products requires better process monitoring and control. The existing sensing techniques for laser powder bed fusion (LPBF) are limited by low spatial and temporal resolutions, as well as the accessibility of sensors. A new process monitoring framework that includes novel physics-based compressive sensing (PBCS) and physics-constrained dictionary learning (PCDL) is proposed. Based on multiphysics models, PBCS allows for the reconstruction of high-fidelity thermofluid latent field from low-fidelity temperature measurements with the seamless integration between models and experiments. In addition, PCDL is developed to reconstruct high-resolution 2D images from the low-resolution ones to further improve sensing efficiency. Machine health states can also be identified from low-fidelity sensor data. PBCS and PCDL mechanisms are demonstrated with thermal and optical images in the LPBF process.
11:20 AM
Defect Prediction on the Base of Thermographic Features in Laser Powder Bed Fusion Utilizing Machine Learning Algorithms: Simon Oster1; Tina Becker1; Philipp Breese1; Nils Scheuschner1; Christiane Maierhofer1; Tobias Fritsch1; Gunther Mohr1; Simon Altenburg1; 1Bundesanstalt für Materialforschung und -prüfung
Avoiding the formation of defects such as keyhole pores is a major challenge for the production of metal parts by Laser Powder Bed Fusion (LPBF). The use of in-situ monitoring by thermographic cameras is a promising approach to detect defects, however the data is hard to analyze by conventional algorithms. Therefore, we investigate the use of Machine Learning (ML) in this study, as it is a suitable tool to model complex processes with many influencing factors. A ML model for defect prediction is created based on features extracted from process thermograms. The porosity information obtained from an x-ray Micro Computed Tomography (µCT) scan is used as reference. Physical characteristics of the keyhole pore formation are incorporated into the model to increase the prediction accuracy. Based on the prediction result, the quality of the input data is inferred and future demands on in-situ monitoring of LPBF processes are derived.
11:40 AM Cancelled
Modeling Strengthening and
Elevated Temperature Properties of Single and Polycrystalline Additively Manufactured Refractory
High Entropy Alloy Microstructures: Anssi Laukkanen1; Tatu Pinomaa1; Matti Lindroos1; Tom Andersson1; Tomi Suhonen1; 1VTT Technical Research Center of Finland
Refractory high-entropy alloys (RHEAs) are an emerging class of materials promising exceptional strength in high operational temperatures. Recent theoretical and computational models propose to better evaluate themechanistic origins of the respective properties, focusing, e.g., in solute to dislocation mobility interactions for thermally-activated glide. Here we investigate the respective mechanisms by using a recently developed machine-learned Gaussian approximated potential (GAP) for the MoNbTaVW RHEA, which can fully capture the chemical complexity of the respective 5-element solid solution. We evaluate and provide insights onto the significance of AM solidification structures regarding these alloys, furthering the alloy-process-property integration and design.
12:00 PM
Numerical Investigation of L-PBF Processing of High Speed Steels to Produce Crack Free Parts: Pravin Kumar1; Akash Bhattacharjee1; Himanshu Nirgudkar1; Surya Ardham1; Pramod Zagade1; BP Gautham1; Gerald Tennyson1; 1TCS Research, Tata Consultancy Services, Pune, India
During L-PBF processing of high-speed steel (HSS), formation of hard and brittle phases lead to cracking of parts under the influence of process induced stresses. Producing crack free HSS components, requires a detailed understanding of interaction among composition, process parameters, and resultant microstructure. In this study, a multiscale framework to investigate L-PBF processing of M2 tool steel is developed. Microstructure based material constitutive model are developed using microstructure statistics predicted by theoretical non-equilibrium solidification models for varying cooling rates, and composition. Part scale simulations are performed to estimate residual stresses and distortion by using the microstructure aware material-constitutive behavior. Numerical trials are conducted to study the influence of variation in alloy composition of M2 tool steel, process parameters on crack susceptibility and final properties. Potential of this in-silico approach in reducing process design space of high speed steels is highlighted.