[Eoas-seminar] [Seminar-announce] Colloquium with Gokberk Kabacaoglu

eoas-seminar at lists.fsu.edu eoas-seminar at lists.fsu.edu
Thu Mar 17 16:47:02 EDT 2022

"Neural Network Based Reduced Model for Stokesian Particulate Flows"

Gökberk Kabacaoğlu
Bilkent University, Ankara, Turkey

NOTE: Please feel free to forward/share this invitation with other groups/disciplines that might be interested in this talk/topic. All are welcome to attend.

Meeting # 942 7359 5552

Mar 23, 2022, Schedule:

* 3:00 to 3:30 PM Eastern Time (US and Canada)
Teatime - Virtual (via Zoom)

* 3:30 to 4:30 PM Eastern Time (US and Canada)
Colloquium - Attend F2F (in 499 DSL) or Virtually (via Zoom)

Stokesian particulate flows describe the hydrodynamics of rigid or deformable particles in the zero Reynolds number regime. Due to highly nonlinear fluid-structure interaction dynamics, moving interfaces, and multiple scales, numerical simulations of such flows are challenging and expensive. I will present our machine-learning-augmented reduced model[1] for fast simulations of such flows. Besides, I will show how the reduced model enables us study optimal microfluidic device design for dense suspensions of deformable particles.

Our goal is to design a deterministic lateral displacement (DLD) device to sort same-size biological cells by their deformability, in particular to sort red blood cells (RBCs) by their viscosity contrast between the fluid in the interior and the exterior of the cells. ADLD device optimized for efficient cell sorting enables rapid medical diagnoses of several diseases such as malaria since infected cells are stiffer than their healthy counterparts. In this context, I will first describe an integral equation formulation[2] that delivers optimal complexity solvers for this type of problems. Despite its excellent theoretical properties, our integral equation solver remains prohibitively expensive for optimization and uncertainty quantification. I will then summarize our efforts to reduce the computational costs, starting from low-resolution discretization, domain truncation, and model reduction.

Model reduction is used to accelerate the action of specific and very expensive nonlinear operators. The final scheme blends ultra low-resolution solvers (who on their own can-not resolve the flow), several regression neural networks, and an operator time-stepping scheme, which we introduced to specifically enable the use of surrogate models. We have used our methodology successfully for flows that are completely different from the flows in the training dataset. This is a joint work with George Biros at the University of Texas at Austin.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.fsu.edu/pipermail/eoas-seminar/attachments/20220317/415a13dd/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: text/calendar
Size: 4788 bytes
Desc: not available
URL: <http://lists.fsu.edu/pipermail/eoas-seminar/attachments/20220317/415a13dd/attachment.ics>
-------------- next part --------------
SC-Seminar-announce mailing list
SC-Seminar-announce at lists.fsu.edu

More information about the Eoas-seminar mailing list