search

okada39 / pinn_cavity

Physics informed neural network (PINN) for cavity flow governed by Navier-Stokes equation.
MIT License
122 stars 30 forks source link

Failure occurs when the Reynolds number increases #1

Open chenchenchen7 opened 4 years ago

chenchenchen7 commented 4 years ago

Dear Sir

I found that when the Reynolds number becomes larger (Re=1000, nu=0.001), the effect is very poor from the drawn image. I don’t understand why this is happening. Could it be that the equation is no longer suitable?

thanks for replying

okada39 commented 4 years ago

Dear chenchenchen7,

We verified that PINN algorithm cannot get an ideal convergence for complex phenomena such as high Reynolds flow and high-dimensional equation. Deep neural networks potentially represent their solutions, but localized ones constrict even though we use quasi-Netwon method.

This problem is a challenge to overcome for implementing PINN application.

AmirhosseinnnKhademi commented 2 years ago

But the problem is that for cases such as Re=10000! the loss values remains the same as the case for Re=400! Why is that? it does not make sense!

AmirhosseinnnKhademi commented 2 years ago

Also, it seems to me that you cannot define separate numbers of training and collocation points for the training! why is that?