Deducing Prior Material Deformation from Simple Mechanics

Description

Process-structure-property linkages suggest an opportunity to deduce processing from behavior. Simple imaging experiments provide rich sources of data. Can we deduce prior deformation?
Scheme: Thin film plasticity

• Deform to some reference strain
• Unload
• Deform to test strain, image
• Deduce reference strain from test-strain image

Relation to AI

• This is an image classification problem
• Lots of good tools for doing AI/ML with images
• Next step is regression
• Deduce arbitrary prior strain, not just which “bucket”
• Include uncertainties

Fundamental idea

• Microstructure encodes history.
• Response to loads is accessible at high resolution and relevant to mechanics.
• Machine learning on images is mature, therefore we can build an ML scheme which ingests image data from mechanical tests, and which infers sample deformation history.

 

Major Accomplishments

• Manuscript on classification (arXiv:1709.8225)
• Provided data to neural-network team
   

Additional Technical Details

Ongoing Challenges

• Data sets required are large, generated from simulations
• Numerous technical options exist for regression schemes, hard to pick

Associated Product(s)

Materials Genome Initiative