Paper Number
DR8                         My Program 

Session
Data-Driven Rheology

Title
Graph neural network in prediction of force chain networks in dense suspensions. Part 2: rheological perspective

Presentation Date and Time
October 16, 2024 (Wednesday) 1:50

Track / Room
Track 6 / Room 501

Authors (Click on name to view author profile)

  1. Aminimajd, Armin (Case Western Reserve University, Macromolecular Science and Engineering)
  2. Maia, Joao (Case Western Reserve University, Macromolecular Science and Engineering)
  3. Singh, Abhinendra (Case Western Reserve University, Macromolecular Science and Engineering)

Author and Affiliation Lines (in printed abstract book)
Armin Aminimajd, Joao Maia and Abhinendra Singh
Macromolecular Science and Engineering, Case Western Reserve University, Cleveland, OH 44106

Speaker / Presenter 
Maia, Joao

Keywords
computational methods; data-driven rheology; dense systems; future of rheology; networks; particles; particualte systems; suspensions

Text of Abstract
The rheological behavior of suspensions, often attributed to the formation of chain contacts arising from particle motion constraints, remains a complex phenomenon influenced by factors such as friction constants, packing fraction, and stiffness. While LF-DEM serves as a valuable tool to understand these influences and predict rheological properties, the computation of interparticle forces for capturing Force Chain Networks (FCN) demands substantial time and energy resources. Leveraging machine learning, particularly Graph Neural Networks (GNN), offers a promising avenue to expedite such predictions. In Part 1, we demonstrated the scalability of GNN in predicting FCN across suspension systems of varying sizes, bidispersity levels, and stresses. Herein, we extend this work by employing deep GNN techniques to interpolate and extrapolate FCN under diverse conditions of friction constants, packing fraction, and stiffness. Our findings suggest that these predictions can be achieved with minimal resource consumption and time by training the model on a limited dataset. This approach holds significant potential for fast understanding of suspension rheology and guiding industrial applications.