Machine Learning-Aided Parametrically Homogenized Crystal Plasticity Model (PHCPM) for Single Crystal Ni-Based Superallo
- PDF / 2,318,858 Bytes
- 16 Pages / 593.972 x 792 pts Page_size
- 27 Downloads / 192 Views
https://doi.org/10.1007/s11837-020-04344-9 Ó 2020 The Minerals, Metals & Materials Society
AUGMENTING PHYSICS-BASED MODELS IN ICME WITH MACHINE LEARNING AND UNCERTAINTY QUANTIFICATION
Machine Learning-Aided Parametrically Homogenized Crystal Plasticity Model (PHCPM) for Single Crystal Ni-Based Superalloys GEORGE WEBER,1 MAXWELL PINZ,2 and SOMNATH GHOSH
3,4
1.—Department of Mechanical Engineering, Johns Hopkins University, Baltimore, USA. 2.—Department of Civil & Systems Engineering, Johns Hopkins University, Baltimore, USA. 3.—Departments of Civil & Systems Engineering, Mechanical Engineering, Materials Science & Engineering, Johns Hopkins University, 3400 N. Charles St., Baltimore, MD 21218, USA. 4.—e-mail: [email protected]
This article establishes a multiscale modeling framework for the parametrically homogenized crystal plasticity model (PHCPM) for single crystal Nibased superalloys. The PHCPMs explicitly incorporate morphological statistics of the c c0 intragranular microstructure in their crystal plasticity constitutive coefficients. They enable highly efficient and accurate calculations for image-based polycrystalline microstructural simulations. The single crystal PHCPM development process involves: (1) construction of statistically equivalent RVEs or SERVEs, (2) image-based modeling with a dislocation-density crystal plasticity model, (3) identification of representative aggregated microstructural parameters, (4) selection of a PHCPM framework and (5) selfconsistent homogenization. Novel machine learning tools are explored at every development phase. Supervised and unsupervised learning methods, such as support vector regression, artificial neural networks, k-means, and symbolic regression, enhanced optimization, model emulation and sensitivity analysis methods are all critical components of the multiscale modeling pipeline. The integration of machine learning tools with physics-based models enables the creation of powerful single crystal constitutive models for polycrystalline simulations.
PHCPM ML ICME RAMP SERVE FIB SEM CPFEM GA GSA KW SSD GND
Nomenclature Parametrically homogenized crystal plasticity model Machine learning Integrated computational materials engineering Representative aggregated microstructural parameter Statistically equivalent representative volume element Focused ion beam Scanning electron microscope Crystal plasticity finite element model Genetic algorithm Global sensitivity analysis Kear-Wilsdorf Statistically stored dislocation Geometrically necessary dislocation
(Received July 13, 2020; accepted August 19, 2020)
SVR ANN
Support vector regression Artificial neural network
INTRODUCTION Machine learning (ML) is firmly within the scientific zeitgeist, and its increasing application in the Integrated Computational Materials Engineering (ICME) discipline has not been an exception. ICME advocates the integration of physicsbased modeling at multiple material scales and emphasizes the links between these scales.1,2 Machine learning is enabling a paradigm shift in ICME, allowin
Data Loading...
