Innovations in Data Engineering and Sciences Materials Design

  • Slides: 26
Download presentation
Innovations in Data Engineering and Sciences: Materials Design, Development & Deployment

Innovations in Data Engineering and Sciences: Materials Design, Development & Deployment

Materials - Manufacturing & Design Advances in both materials and manufacturing sciences have ushered

Materials - Manufacturing & Design Advances in both materials and manufacturing sciences have ushered in a new era of materials innovation and deployment; the central impediment comes from lack of a suitable framework for efficient knowledge exchange between these fields.

The Opportunity in Materials Innovation • Scale-up of Innovation: Digital Knowledge Systems • Accelerated

The Opportunity in Materials Innovation • Scale-up of Innovation: Digital Knowledge Systems • Accelerated Innovation: Templated Workflows • Culture Change: e-Collaborations

 • Building on Georgia Tech’s Leadership in Materials Innovation Infrastructure… • • •

• Building on Georgia Tech’s Leadership in Materials Innovation Infrastructure… • • • Innovation Initiatives & Shared Resources • • • e-Collaboration Utilize open source and open access data/code repositories Facilitate crossdisciplinary team discussions and annotations of intermediate results Manage workflows and identify best practices Decision support for future investments with high ROI Data Management Capture Storage Aggregation Sharing Protocols Knowledge databases MD 3 will facilitate exploration of open source and commercial data sciences methods within the materials innovation ecosystem Data Analytics Mine embedded high value information via: • Filtering • Data fusion • Uncertainty analyses • Statistical analyses • Dimensionality reduction • Pattern recognition • Regression analysis • Machine learning • Statistical learning

Innovation Ecosystem of IDEAS: MD 3 Small companies Design optimization and PLM OEMs Data

Innovation Ecosystem of IDEAS: MD 3 Small companies Design optimization and PLM OEMs Data services vendors Materials suppliers Materials R&D Emphasis is placed on workforce development in materials data sciences and informatics, and integration within the ecosystem 5

Leveraged Partnership MATIN 24 -story, 695, 000 SF private and public development

Leveraged Partnership MATIN 24 -story, 695, 000 SF private and public development

Core Materials Knowledge: Process-Structure-Property (PSP) Linkages Process Space Structure Space Properties Space • Each

Core Materials Knowledge: Process-Structure-Property (PSP) Linkages Process Space Structure Space Properties Space • Each material structure is associated with only one value of a property • Each hybrid process can be depicted as a pathline in the structure space • If formulated as reliable reduced-order linkages, it will be possible to address most of the MD 3 challenges

Templated Workflows for Hierarchical PSP Linkages Step 1: Convert microstructure image into a digital

Templated Workflows for Hierarchical PSP Linkages Step 1: Convert microstructure image into a digital signal Step 2: Compute n-point spatial correlations (capture important features identified by known physics) Step 3: Obtain low dimensional structure measures using dimensionality reduction (e. g. , principal component analyses) Step 4: Utilize structure measures in learning PSP linkages Polymer Chains Simulation Data

n-Point Spatial Correlations • Spatial correlations capture all of the salient measures of the

n-Point Spatial Correlations • Spatial correlations capture all of the salient measures of the microstructure • Efficient codes for computing them are now available through Py. MKS code repository %50 %0

Principal Component Analysis (PCA) Data-driven (objective) dimensionality reduction based on capturing the maximum variance

Principal Component Analysis (PCA) Data-driven (objective) dimensionality reduction based on capturing the maximum variance in the dataset in the minimum number of orthogonal components Original Axes y Principal Axes p 1 p 1 p 2 PC 1 Reduced Axis p 2 x Theory 10 PC 2 PC 3 4. 552 -1. 182 0. 058 2. 176 1. 243 0. 825 -3. 421 -1. 06 0. 621 -5. 104 0. 304 -0. 147 0. 475 -1. 252 -0. 074 4. 939 0. 628 -0. 598 -1. 076 0. 988 0. 389 -2. 541 0. 332 -1. 073

Application: Vendor Qualification • • Compile legacy microstructures and their variances into database for

Application: Vendor Qualification • • Compile legacy microstructures and their variances into database for quantitative comparison Build low cost, reliable PS-P models using available data from experiments and simulation Establish data-driven decision system to qualify acceptable/not acceptable performance with metrics for uncertainty Design experimental and/or modeling protocols to rapidly acquire data need to enhance decision system m 5 – sample 3 m 1 1 – sample 8 2 m 8 – sample 3 18 Acceptable Microstructure Databases Acceptable Structure-Property Databases

Application: Phase-Field Simulations Initial Microstructure Ag-Al-Cu System (Ternary Eutectoid Alloys) nal tion rec Di

Application: Phase-Field Simulations Initial Microstructure Ag-Al-Cu System (Ternary Eutectoid Alloys) nal tion rec Di difica li So Microstructure Evolution Final Microstructure Johannes Hötzer, Marcus Jainta, Philipp Steinmetz, Britta Nestler, Anne Dennstedt, Amber Genau, Martin Bauer, Harald Köstler, Ulrich Rüde, Large scale phase-field simulations of directional ternary eutectic solidification, Acta Materialia, 2016 Final microstructure

Process-Structure Linkages Distribution of final microstructures with respect to boundary velocity Distribution of microstructures

Process-Structure Linkages Distribution of final microstructures with respect to boundary velocity Distribution of microstructures with similar velocities but different concentrations Process-structure linkages can be mined to predict microstructures for new combinations of velocity and concentration values.

Application: GB Atomic Structure Tschopp et al. , IMMI, 2015: 106 Datasets; Energy-minimized Al

Application: GB Atomic Structure Tschopp et al. , IMMI, 2015: 106 Datasets; Energy-minimized Al GBs; Σ 3, 5, 9, 11, 13; Inclination angle 0 -90° Workflow: Identify GB atoms; Calculate pair correlation functions; Predict GB energy using PCA regression

GB Structure-Energy Linkages First Place Winner in the National Data Challenge: $25, 000 Cash

GB Structure-Energy Linkages First Place Winner in the National Data Challenge: $25, 000 Cash Prize

Stress Fields in Polycrystals • 45 x 45 Microstructure. Each color represents a distinct

Stress Fields in Polycrystals • 45 x 45 Microstructure. Each color represents a distinct crystal lattice orientation randomly selected from cubic FZ. • FEM prediction: 3 minutes with 16 processors on a supercomputer • MKS prediction: 30 seconds with only 1 processor on a standard desktop computer

High Throughput Experiments for PSP Linkages High throughput prototyping of high value microstructures through

High Throughput Experiments for PSP Linkages High throughput prototyping of high value microstructures through controlled thermal and/or mechanical gradients Instrumented indentation is capable of providing quantitative stress-strain responses at length scales ranging from 50 nms to 500 microns Undeformed Deformed 24% height reduction Strain Map from DIC

A new GT Center: Innovations in Data Engineering and Science for Materials Design, Development,

A new GT Center: Innovations in Data Engineering and Science for Materials Design, Development, and Deployment Use-inspired research in close collaboration with industry and national lab partners Cyberinfrastructure to enhance productivity and management of cross-disciplinary ecollaborations High throughput multiscale measurements of structure and response(s) informed by models IDEAS: MD 3 Informatics and data analytics for efficient integration of high value information across hierarchical scales Innovative synthesis/processing techniques with control of hierarchical structure Physics-based multiscale models informed and validated by measurements Uncertainty management and robust design in multiscale material systems

Value Proposition - I • Educate current workforce in emerging concepts and paradigms in

Value Proposition - I • Educate current workforce in emerging concepts and paradigms in data-driven MD 3 Ø Newsletters, seminars, workshops, tutorials, webinars, MOOCs, graduate theses, and an annual conference Ø Class projects in Materials Informatics course sequence Ø Focused (customized) support for your specific needs Ø Access to a new cadre of skilled human resources meeting industry needs

Value Proposition - IIa • Equip your organization with customized modern data science tools

Value Proposition - IIa • Equip your organization with customized modern data science tools and e-collaboration platforms Ø Access to unique bleeding-edge research software with support § Py. MKS (hierarchical materials informatics) § Py. DEM (Inductive design for MD 3) § Py. CAC (Coarse-grained atomistic FEM) § SPIN (Spherical indentation stress-strain) § CRYSP (Crystal plasticity spectral database) § Selected software packages from Sandia, ORNL, etc.

Value Proposition - IIb • Increased exposure of your organization to modern data science

Value Proposition - IIb • Increased exposure of your organization to modern data science tools and ecollaboration platforms for MD 3 Ø MATIN e-collaboration platform built on HUBzero with enhanced functionality § Automated (or semi-automated) data ingest addressing legacy and new data § Curated workflows for analytics § Publication of intermediate research § Curated materials knowledge § Expertise matching § Team/project management

Value Proposition - III • Unique networking opportunities for win-win partnerships at the nexus

Value Proposition - III • Unique networking opportunities for win-win partnerships at the nexus of materials science, manufacturing, and data science Ø Identify new market opportunities for commercial data/compute services Ø Seamless connectivity between the OEMs and their supply chains Ø Domain experts in emerging areas related to MD 3

Value Proposition - IV • Training in novel high throughput strategies for rapid exploration

Value Proposition - IV • Training in novel high throughput strategies for rapid exploration of design spaces in MD 3 Ø Novel experimental protocols based on high throughput assays involving sample libraries with small volumes of materials Ø Novel testbeds to evaluate and explore new concepts of interest to your organization in sandboxed environments

Membership Levels • Regular Member: $5, 000 Ø Three participants at annual IDEAS: MD

Membership Levels • Regular Member: $5, 000 Ø Three participants at annual IDEAS: MD 3 conference • Affiliate Member: $1, 500 Ø User access to MATIN Ø Allows a one year trial Ø Access to networking Ø One participant at annual meetings with GT faculty IDEAS: MD 3 conference and students Ø User access to MATIN Ø Support of 20 hours Ø Support of 10 hours Ø One representative permitted to attend meetings of the Member Advisory Board (no voting privileges)

Membership Levels • Full Voting Member: $10, 000 Ø Five participants at annual IDEAS:

Membership Levels • Full Voting Member: $10, 000 Ø Five participants at annual IDEAS: MD 3 conference Ø User access to MATIN Ø Access to networking meetings with GT faculty and students Ø Access to limited office space Ø Support of 40 hours Ø One representative on Member Advisory Board (voting privileges)

Demos • MATIN e-Collaboration Platform: Dr. Aleks Blekh • Experimental Lab Automation (ELA): Ali

Demos • MATIN e-Collaboration Platform: Dr. Aleks Blekh • Experimental Lab Automation (ELA): Ali Khosravani and Dr. Soumya Mohan • Data Analytics using Py. MKS: Ahmet Cecen and Noah Paulson • Collaborative Projects using MATIN: Dr. Andrew Medford and Dr. Evdokia Popova