2023 Annual International Solid Freeform Fabrication Symposium (SFF Symp 2023): Data Analytics: Application to General Additive Manufacturing
Program Organizers: Joseph Beaman, University of Texas at Austin
Tuesday 8:15 AM
August 15, 2023
Room: 415 AB
Location: Hilton Austin
Session Chair: Nicholas Meisel, Pennsylvania State University
8:15 AM Cancelled
Nowcasting Melt Pool Dynamics using Transformer on In-situ Monitoring Images: Shu Wan1; Hyunwoong Ko1; 1Arizona State University
Accurate understanding and prediction of melt-pool dynamics are crucial for Additive Manufacturing (AM) processes, as they significantly impact final products’ quality and mechanical properties. In this paper, we propose a novel method, Melt-pool Generative Pre-trained Transformer (MeltpoolGPT), for predicting the next melt pools using a GPT-based architecture specifically designed for video frame predictions. Our approach captures spatio-temporal dependencies between melt pools and learns the underlying physical dynamics governing laser-powder-bed-fusion processes. We evaluate MeltpoolGPT on melt-pool image data acquired from the AM Metrology Testbed at the National Institute of Standards and Technology, and design a series of experiments to assess the accuracy and stability of its predictions. Our experiments indicate that MeltpoolGPT achieves high accuracy in predicting the next melt-pool images, outperforming other state-of-the-art methods. This work presents a first step in predicting future melt pools, a largely unexplored area with immense potential to benefit real-time monitoring and control in AM significantly.
8:35 AM
Dreaming of Data: Examining Data Augmentation for Machine Learning in Additive Manufacturing: Glen Williams1; Martha Baldwin2; Timothy Simpson1; Nicholas Meisel1; Christopher McComb2; 1Pennsylvania State University; 2Carnegie Mellon University
The data generated during additive manufacturing (AM) practice can be used to train machine learning (ML) tools to reduce defects, optimize mechanical properties, or increase efficiency. In addition to the size of the repository, emerging research shows that other characteristics of the data also impact suitability for AM-ML application. What should be done in cases for which the data in too small, too homogeneous, or otherwise insufficient? Data augmentation techniques present a solution, offering automated methods for increasing the quality of data. However, many of these techniques were developed for machine vision tasks, and hence their suitability for AM data has not been verified. In this study, several data augmentation techniques are applied to synthetic design repositories to characterize if and to what degree they enhance their performance as ML training sets. We discuss the comparative advantage of these data augmentation techniques across several canonical AM-ML tasks.
8:55 AM
A Framework for Physics-guided Machine Learning to Extract and Transfer Process-structure-property Knowledge in Additive Manufacturing: Hyunwoong Ko1; Fatemeh Elhambakhsh1; 1Arizona State University
Emerging research in Additive Manufacturing (AM) seeks to pursue Machine Learning (ML) that can improve the understanding of Process-structure-property (PSP) causality. To address the challenge, we provide a novel framework for physics-guided ML to extract and transfer PSP knowledge. The framework first uses an approach guided by physics knowledge graphs to generate the requirements for predictive PSP analytics. Then, the framework uses physics-informed ML to construct new PSP knowledge. The study enables ML to systematically couple physics knowledge with the versatility of cyber-physical AM data in PSP analytics. This study also provides a foundational basis for AM to synergically merge newfound knowledge about PSP from data with a priori physics knowledge. The framework continuously updates coupled PSP linkages to improve the understanding of dynamic AM processes. The continuous PSP learning accumulates structured newfound PSP knowledge in iterations for future ML and proactive control decisions.
9:15 AM
Additive Manufacturing (AM) Lattice Segmentation and Analysis Enabled through Deep Learning: Michael Juhasz1; Gabe Guss1; J. B. Forien1; Nick Calta1; 1Lawrence Livermore National Laboratory
Ex-situ computed tomography (CT) analysis of Additive Manufacturing (AM) produced parts is commonplace as a means of Non-Destructive Evaluation (NDE) quality assurance. Most CT examinations focus on porosity, both from keyholing or entrained gas. With the recent acceleration in image processing enabled through Deep Learning/Machine Learning (ML/DL), this presentation suggests expanding CT analysis of AM parts to extend beyond porosity analysis to encompass the study of other requirement-driven, critical geometries. This was applied to AM produced lattices which underwent CT and were subsequently segmented into component pieces. These segmented components were then registered to in-situ diagnostic signals for comparison where dependence and correlation was assessed, and it is those results which will be presented. This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344.
9:35 AM
Machine-Learning-Driven Digital Twin Construction for Additive Manufacturing: A Review: Fatemeh Elhambakhsh1; Hyunwoong Ko1; 1Arizona State University
Machine Learning (ML) on high-value data, both from cyber and physical systems, has significant potential for constructing a novel Digital Twin (DT) of Additive Manufacturing (AM). However, the use of ML has been largely hindered in the AM DT construction due to the limited understanding of the potential. To address the limitation, in this study, we thoroughly identify ML’s capability and newfound opportunities driven by ML in the DT construction for AM. This study reviews AM DTs’ key features and emerging applications, and state-of-the-art ML methods for the DT construction. This study also discusses open issues and outlooks on the future directions of the ML-driven DT construction for AM. This study helps maximize ML on cyber-physical AM data in automatically constructing spatiotemporally scalable DTs that improve the understanding of physical phenomena and control decisions in AM.
9:55 AM Break
10:25 AM
PRISM: Process Parameter Optimization for Selective Manufacturing: Anthony Garland1; Dale Cillessen1; Kaitlynn Conway1; Johnson Kyle1; Brad Boyce1; Jay Carroll1; 1Sandia National Laboratories
The process of selecting optimal process parameters for additive manufacturing (AM) of a new material can be a challenging task, particularly for materials with limited information in the literature. This study describes an iterative approach that uses machine learning algorithms to predict printability and the target material property when given a set of process parameters. These ML algorithms when combined with optimal design of experiments, enable the selection of optimal process parameters for additive manufacturing (AM) of new materials. The results demonstrate the effectiveness of our approach. This study provides a framework for future research in the selection of optimal process parameters for manufacturing processes and highlights the potential of machine learning in optimizing materials design and manufacturing processes.
10:45 AM
Towards a FAIR Knowledge Management System for Additive Manufacturing: Shengyen Li1; Yan Lu1; Paul Witherell1; 1National Institute of Standards and Tecnhology
Additive manufacturing builds near-net-shape parts using high volume data for designing, operating, and certifying processes and products. To cost-effectively mature AM technology, a FAIR (Findable, Accessible, Interoperable, and Reusable) infrastructure is developed that enables integrations among machines, analytics, and computer tools. This infrastructure includes data models, which is developed following ASTM standards defining key terminologies in a structure for archiving AM related data. A selected dataset, including AM building files, in-situ sensing data, and measuring results, are used to validate the data models. This infrastructure also provides API tools to manage raw data and metadata for statistical analyses to identify the variability of the processes and assess the data quality. The results from this analytical process assist the following sensitivity analyses for the design of experiments and predictive model developments. This presentation will share a case study of identifying the metrology and analytics gaps and opportunities using this informatics-based framework.
11:05 AM
Usage of Unconventional Data Sources for Market Intelligence (MI) in the Field of Additive Manufacturing (AM) - Expert Networks, Technology Territories and Trends: Stephan Ziegler1; 1RWTH Aachen University - Digital Additive Production DAP
The geographic expansion of the markets for AM increasingly confronts companies with greater competitiveness due to the globalization. In addition, market participants are facing rapid changes in the business environment - due to new information and communication technologies. Companies only have a chance to hold their market position if they quickly adopt market changes. Therefore, the decision-making process needs to be accelerated by on-demand information provision. MI offers one possibility to meet these requirements, but typically based on external unstructured data for market and competitive evaluation, which makes it cost and time consuming.A specific investigation of such data sources related to MI for systematic use within the AM markets is being carried out. For this purpose, different data sources (e.g. LinkedIn) will be identified, analysed with focus on information synthesis using text mining and their suitability for the evaluation of expert networks, technology territories and trends be presented.
11:25 AM
Virtual Inspection of Advanced Manufacturing via Digital Twins: Brian Giera1; Brian Weston1; Ziad Ammar1; Seth Watts1; Haichao Miao1; 1Lawrence Livermore National Laboratory
A digital twin (DT) is an amalgam of physics-based and/or data driven models that describe a physical system. In additive manufacturing (AM), inputs/outputs of digital and physical twins are identical. Like many, LLNL’s approach requires refining DTs with data to better capture behavior of its physical twin via advanced analytical techniques. As such, a DT’s evolving parameter set can inform of machine health and aging behavior, providing actionable insights on lifetime performance. A suite of DTs that capture all fabrication and inspection platforms of a given AM process can accelerate production for qualifying parts at scale with minimized and quantified defects. This talk will walk through examples of how we are leveraging data from integrated pairs of real and digital twins of inspection and fabrication platforms to become more flexible and agile.This work was performed under the auspices of the U.S. DOE by LLNL under Contract DE-AC52-07NA27344, LLNL-ABS-820420.