Freie Universität Berlin - Blueprints for automated material discovery using artificial intelligence
HKUST-1 robot using a genetic algorithm to optimize for high crystallinity and high BET surface across 9 design variables
https://www.materialscloud.org/work/tools/sycofinder
Integrated Computational Materials Engineering
VTT ProperTune as a materials acceleration platform (MAP).
Materials discovery and design:
- HEADFORE, HIERARCH (AoF)
- HIDDEN (H2020-BAT)
- AIMS + synbio activities (VTT) Materials optimization, inference, inverse problems
- COMPASSCO2 (H2020-SPIRE)
- ENTENTE (H202-EURATOM)
- EUROfusion (HEU-EUROfusion)
- ACHIEF (H2020-SPIRE) Physics- and data-driven hybrids
- BF-ISA, BF-AVE (BF) Data Analysis, Surrogates, ROM
- GREENY, CORTOOLS etc. (EIT Raw Materials)
Discovery of new electrolytes and electrode materials
Scholar, LinkedIn
Anamoly detection, explainable AI
Materials Zone - From Materials Data to AI Accelerated Results, Fast!
Similar to another revolution, customer relationship management (CRM), is the revolution of materials informatics platforms (MIPs), also referred to as materials acceleration platforms (MAPs).
Data ingestor format
Interactive exploratory data analysis (EDA): correlation matrices, histogram exploration
Showcase of platform using an open-source database http://www.perovskitedatabase.com/
Proven Use-cases
- Innovation - R&D Acceleration - Less/Shorter Cycles
- Sales tool - Find optimal formulations rapidly, accurately
- Supply Chain - Find cheaper, better, more reliable substitutes
- Scale-up - from lab to mass-production, Faster
- Manufacturing - Q.C. - stop bad batches early
- Cross Functional - Supply Chain, R&D, Manufacturing, Business Proven Domains
- Polymers/Composites, Photovoltaics, Building Materials, Nanotech
- Health and Wellness, Batteries, Hydrogen, Metals, Alloys
- 3D Printing, 2D Printing, Surfaces, Films, Packaging, Chemicals
Compare with Schrodinger. iterative Qubit Coupled Cluster
Formulation optimization, liquid formulations (e.g. coatings)
"Not so much an algorithm as a process"
https://github.com/nasa/pretrained-microscopy-models
https://scholar.google.com/citations?user=ofwM5BMAAAAJ&hl
Future of CAMD in multi-fidelity and multi-objective active learning.
Adopting a more event-based model, keeping track of the state of the model. Many-to-one and one-to-one properties like phase stability.
