Hi everyone, I’m working on solving the 2D Navier-Stokes equations in a straight channel using PINNs. While the training loss is low (~1e-6), the outlet velocity profile is linear instead of parabolic. I've enforced no-slip at walls and uniform inlet velocity. Has anyone faced this issue or can offer suggestions?
Hi Kinshuk Shekhar ,
The preferred solution will be to check out the model weight in your case.
Possible problems might be due to incorrect boundary conditions, just wanted to know if you have incorporated Neumann boundary conditions in your study, and at what Reynolds' number you are working.
From my experience struggling with NVIDIA Modulus and several codes from classics (https://github.com/maziarraissi/PINNs) :) I would say without detailed description of the model and how you set it within the framework which you used it's impossible to troubleshoot anything. It can be literally anything.