Materials Informatics Overview
4 likes1,768 views
The document provides an overview of materials informatics and the Materials Genome Initiative. It discusses how materials informatics uses data-driven approaches and techniques from fields like signal processing, machine learning and statistics to generate structure-property-processing linkages from materials science data and improve understanding of materials behavior. This includes extracting features from materials microstructure, using statistical analysis and data mining to discover relationships and create predictive models, and evaluating how knowledge has improved.
Materials Informatics Overview
- 1. + Materials Informatics Overview Tony Fast NIST Workshop – Monday, January 13, 2014
- 2. + The Materials Genome Initiative Experiment Digital Data Simulation MGI places a new focus on how materials generators and materials data analysts create and ingest new and legacy information.
- 3. + Materials Science Knowledge Structure Process Property Information is generated with the goal of improving the knowledge of structure-property-processing relationships.
- 4. + An Applied Representation of Materials Information S c a l e Homogenization . Localization . Time Physics based models, via either simulation or experiment, are designed and refined to generate structure-response information that will either support or challenge the current knowledge of the material behavior.
- 5. + An Applied Representation of Materials Information S c a l e Homogenization . Localization . Models generate relationships between the structure and its effective response (bottom-up), its local response (topdown), or its change during processing. Time The responses or changes are controlled by the mesoscale arrangement of the material features. The materials structure is the independent variable.
- 6. + Some Spatial Material Features Most information generated is spatial & really expensive. Volume Variety Velocity
- 7. + A lot of the spatial information is ignored CT information Top view Cut out a square, its easier.
- 8. + Microstructure Informatics n Microstructure informatics is an emerging data-driven approach to generating structure-property-processing linkages for materials science information. n Microstructure informatics appropriates ideas from signal processing, machine learning, computer science, statistics, algorithms, and visualization to address emerging and legacy challenges in pushing the knowledge of materials science further.
- 9. + Microstructure Informatics INTELLIGENT DESIGN OF EXPERIMENTS PHYSICS BASED MODELS SIMULATION | EXPERIMENT MICROSTRUCTURE (MATERIAL) SIGNAL MODULES ADVANCED & OBJECTIVE STATISTICAL MODULES DATA MINING MODULES VALUE ASSESSMENT Scrape the relevant data and metadata about the structure, responses, and structure changes from any available simulated or experimental models.
- 10. + Microstructure Informatics INTELLIGENT DESIGN OF EXPERIMENTS PHYSICS BASED MODELS SIMULATION | EXPERIMENT MICROSTRUCTURE (MATERIAL) SIGNAL MODULES ADVANCED & OBJECTIVE STATISTICAL MODULES DATA MINING MODULES VALUE ASSESSMENT Eke out the desired features & encode them into signals that can be analyzed.
- 11. + , Grains Grain Boundaries, & Grain Orientations
- 12. + Fiber Centroids in a Massive 3-D Image
- 13. + Heterogeneous Signals in Polycrystals
- 14. + Microstructure Informatics INTELLIGENT DESIGN OF EXPERIMENTS PHYSICS BASED MODELS SIMULATION | EXPERIMENT MICROSTRUCTURE (MATERIAL) SIGNAL MODULES ADVANCED & OBJECTIVE STATISTICAL MODULES DATA MINING MODULES VALUE ASSESSMENT Use algorithms and image processing to extract statistics from the material structure to use as the independent variable in the informatics process.
- 15. + Grain size, Grain Faces, Number of Grains, Mean Curvature, & Nearest Grain Analysis
- 17. + Vector Resolved Spatial Statistics
- 18. + Microstructure Informatics INTELLIGENT DESIGN OF EXPERIMENTS PHYSICS BASED MODELS SIMULATION | EXPERIMENT MICROSTRUCTURE (MATERIAL) SIGNAL MODULES ADVANCED & OBJECTIVE STATISTICAL MODULES DATA MINING MODULES VALUE ASSESSMENT Numerical methods, machine learning, and new models to create structure-propertyprocessing linkages.
- 19. + Data mining applications & the goal of the workshop n Homogenization – Improved bottom-up linkages using improved feature detection, richer datasets, & better statistical descriptors. n Localization – “How can I execute a model on a new material structure faster and sacrifice precision a tiny bit?” n Structure-Structure – Quantitative comparison between materials with different structures, but similar ontologies. We will solve localization problems today, homogenization and structure quantification are tomorrow."
- 20. + Microstructure Informatics INTELLIGENT DESIGN OF EXPERIMENTS PHYSICS BASED MODELS SIMULATION | EXPERIMENT MICROSTRUCTURE (MATERIAL) SIGNAL MODULES ADVANCED & OBJECTIVE STATISTICAL MODULES DATA MINING MODULES VALUE ASSESSMENT How much did the knowledge improve? Is new data needed? Is a better mining technique available? Can better statistics be extracted? Can another feature be included?
- 21. + Success Stories in Microstructure Informatics n Homogenization n n Localization n n n Improved regression models for the diffusivity in fuel cells Meta-models for spinodal decomposition Meta-models for highly nonlinear elastic, plastic, and thermomechanical responses Structure-Structure n n n Quantitative comparison between heat treated a-b experimental Titanium datasets. Degree of crystallization in Polymer Molecular Dynamics simulation. Model verification in Molecular Dynamics simulations.
- 22. + Materials Knowledge System Overview n Localization is provides a spatially resolved response for a particular material structure FEM" ε=5e-4" h ps = ∑∑ ath ms+t t h
- 23. Any Model + Materials Knowledge System Overview Generalized INPUT Control" OUTPUT h ps = ∑∑ ath ms+t t h The MKS design filters that capture the effect of the local arrangement of the microstructure on the response. The filters are learned from physics based models and can only be as accurate as the model never better.
- 24. + Applications of Localization n Model scale is intractable n Fast, scalable, computationally linkages are necessary efficient top-down
- 25. + Information & Knowledge Microstructure Signal Response Signal Same Size Under a set of control parameters and boundary conditions, the arrangement of the features described by the microstructure signal can be connected to the final response the arrangement
- 26. + Information & Knowledge Microstructure Signal Response Signal Regression transforms information to knowledge in the form of influence coefficients
- 27. + The Influence Coefficients n Contain knowledge of the physics expressed by the material information n Any assumptions, or uncertainty, is propagated in the influence coefficients. n Originally devised from Kroner’s on heterogeneous medium n The are filters that contain the physics of the spatial interaction with the spatial arrangement of features n n Symmetric-first derivative of the Green’s function Relates to perturbation theory n Have fading memory n Can be scaled. h ps = ∑∑ ath ms+t t h Convolution Relationship
- 28. + Image Filtering h(u, v ) f (x, y ) h =1 g = h∗ f g (x, y )
- 29. + Image Filtering - Blurring h(u, v ) f (x, y ) h(u, v) = ⎡0 01 0 0⎤ ⎢0 1 1 1 0⎥ ⎢ ⎥ ⎢1 1 1 1 1 ⎥ ⎢ ⎥ 0 1 1 1 0⎥ ⎢ ⎢0 01 0 0⎥ ⎣ ⎦ g = h∗ f g (x, y )
- 30. + Image Filtering - Embossing h(u, v ) f (x, y ) h(u, v) = ⎡− 1 − 1 0⎤ ⎢− 1 0 1⎥ ⎢ ⎥ ⎢ 0 1 1⎥ ⎣ ⎦ g (x, y ) Filtering modifies a pixel at (x,y) by some function of the local g = h ∗ f by h neighorhood defined
- 31. + Generating Knowledge – A workflow 1. Gather or generate microstructure and spatial response information 2. Extract and encode the feature of the microstructure 3. Calibrate the Influence Coefficients 1. 2. Choose an encoding Choose a calibration set 4. Fourier transform of microstructure and response signal Calibrate in the Fourier space 5. Convert influence coefficients to the real space 3. 4. Validate the Influence Coefficients
- 32. + Core elements of the Materials Knowledge System n What we need to know n Methods to determine independent and dependent variables Linear regression n Prior knowledge about your information n n What we need to use n Fast Fourier Transforms n Linear Regression n Numerical Methods to generate data
- 33. + Fourier Transforms of a Convolution n The Fourier space decouples the spatial dependencies n The influence coefficients are calibrated in the Fourier space because the initially it appears to simplify the problem.
- 34. + Topology of the Influence Coefficients Fading Memory a 63 t Influence scaling easy because of the fading memory and scale better than most models.
- 35. + Application: Spinodal Decomposition (1) • From an initial starting structure, ONE set of influence coefficients can be used to evolve the material structure" Time Derivative" MSE Error"
- 36. + Application: Spinodal Decomposition (2) Time Derivative" MSE Error"
- 37. + Application: High contrast elasticity The MKS is a scalable, parallel meta-model that learns from physics based models to enable rapid simulation at a cost in accuracy. N2 vs. Nlog(N) complexity It learns top-down localization relationships to extra extreme value events and enables multiscale integration. OTHER APPLICATIONS" Spinodal Decomposition, Grain Coarsening, " Thermo-mechanical, Polycrystalline
- 38. + On to the next one. Have Fun!