Distributed heterogeneous compute infrastructure for the study of additive manufacturing systems
- PDF / 1,520,725 Bytes
- 9 Pages / 432 x 648 pts Page_size
- 30 Downloads / 162 Views
MRS Advances © 2020 Materials Research Society DOI: 10.1557/adv.2020.103
Distributed heterogeneous compute infrastructure for the study of additive manufacturing systems Mathew Thomas1, Malachi Schram1, Kevin Fox1, Jan Strube1, Noah S. Oblath1, Robert Rallo1, Zachary C. Kennedy1, Tamas Varga1, Anil K. Battu1, Christopher A. Barrett1 1
Pacific Northwest National Laboratory, Richland, Washington, U.S.A.
ABSTRACT We present the current status of a scalable computing framework to address the need of the multidisciplinary effort to study chemical dynamics. Specifically, we are enabling scientists to process and store experimental data, run large-scale computationally expensive high-fidelity physical simulations, and analyze these results using state-of-the-art data analytics, machine learning, and uncertainty quantification methods using heterogeneous computing resources. We present the results of this framework on a single metadata-driven workflow to accelerate an additive manufacturing use-case.
INTRODUCTION The Chemical Dynamics Initiative (CDI) at Pacific Northwest National Laboratory (PNNL) is developing a capability to predict the temporal evolution of complex chemical systems in real-world environments. Developing this capability requires a holistic and concerted effort composed of data science, theory, experiment, and simulations. The initiative has multiple use-cases targeting different research areas. This paper focuses on the underlying computing infrastructure developed by this initiative to accelerate and automate computational workflows. As an illustrative example, we outline an Additive Manufacturing (AM) build orientation classification workflow based on the analysis of experimental X-ray computed tomography (XCT) data that requires large CPU and GPU processing resources. The ability to determine the build orientation of an AM-produced part using machine learning (ML) is useful for forensic analysis and to gain confidence in the development of generative ML for use as a forward model of the AM process. AM processes often generate parts with anisotropic performance characteristics that correlate strongly with their orientation during their construction (i.e. referred to here as build orientation).
Downloaded from https://www.cambridge.org/core. Rice University, on 28 Apr 2020 at 07:51:30, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1557/adv.2020.103
This paper is organized as follows. Section II provides an overview of the computing infrastructure that includes the virtualized service cluster, the use of the DIRAC interware, and the integration of the PNNL heterogeneous computing resources. Section III is a summary of the AM build orientation classification workflow. Section IV highlights the metrics from the workflow. Finally, Section V presents the conclusion and perspectives of the work presented and a potential path forward. CHEMICAL DYNAMICS COMPUTING INFRASTRUCTURE WORKFLOW Virtualized service cluster The computing infrastr
Data Loading...
