raocp / pinn-elastodynamics Goto Github PK

Dear Ms.Rao

I tried to run your script and it seems that there are missing some methods : train_bgfs_dist and train_bgfs_part in the code. Is it possible to get an update?
Thanks for this very interesting work !!

Nicolas

Dear Mr. Rao, Your work has inspired me a lot, but the code of the example of semi-infinite space has not been uploaded, I would like to ask whether it can be uploaded?

Hello Mr. Rao,
thank you for your nice work, I found it quite inspiring and close to the topic that I am working on. however, the way that you applied Neumann boundary condition is a bit vague to me. Generally speaking, we know that for Dirichlet BC the predefined solution just satisfies the values on the boundaries by having D=0 and T takes the values of Tbc (T = Tbc + D * Tnn). For the Neumann boundary condition (gradT . n = Nbc) the derivatives of T are specified while the value of T is not given but it gets some values that might not be correct. Could you please explain a bit more how the mixed-variables output solved this issue?

best regards,

Alborz

Dear authors,
I read your interesting paper arXiv:2006.08472v1
But the link to the code https://github.com/Raocp/PINN-elastodynamics is empty.
Can you update the repository, please?

Hi Chengping,

Thank you for sharing a very interesting paper and the code!

I have a issue when running the code, and hope that you may help me out.

The problem is that, if I didn't use the existed file "uv_NN_10s.pickle" to initialize the network, and instead using the function initialize_NN to intialize the weights and biases, the training will fail after 70 iterations. Any insights for how to properly intialize the nn?

Thank yo so much in advance.

Regards

Hello,
In the infinite wave field,I found that there are only one DNN to generate (u, v, ut, vt, s_11, s_12, s_22). there is no code that mean the pretrained IC/BC DNN or distance function DNN. And in the confined boundary, the model prediction (u, v, ut, vt, s_11, s_12, s_22) only compute with $N^{(3)}:net_{uv}$ and don't realted with $N^{(1)}:net_{part}$ and $N^{(2)}:net_{dist}$. But in the paper, i understand the model prediction should be the $N^{(3)}*N^{(2)} + N^{(1)}$.I don't know if it is right.
Best regards,
Dong

Dear Mr. Rao,
Is it possible to upload the code? This page is empty.