The training process struggles to converge when using CFD flow field data that I've trained myself.

[HCwinwinwin]

I performed a simulation of two-dimensional flow around a cylinder using ANSYS Fluent, extracted transient simulation data to form a dataset, and then selected the vortex shedding region behind the cylinder for training in your model. This resulted in the aforementioned outcome.
Ask for an answer, thank you.

[Shengfeng233]

Well, this looks like your data is wrong. You can check if your data is non-dimensional data, since the underlying equations presented in the code is non-dimensional NS equations. PIus, you can also check whether your CFD simulation is correct. Best, Shengfeng At 2024-02-27 21:52:25, "HCwinwinwin" ***@***.***> wrote: image.png (view on web) I performed a simulation of two-dimensional flow around a cylinder using ANSYS Fluent, extracted transient simulation data to form a dataset, and then selected the vortex shedding region behind the cylinder for training in your model. This resulted in the aforementioned outcome. Ask for an answer, thank you. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[HCwinwinwin]

Thanks！
Before inputting the data, I processed it according to the steps mentioned, and it seems like there shouldn't be any issues.

The results from the simulation in Fluent seem to be normal.Is it possible that the grid size of 0.5mm in my vortex shedding region is not accurate enough, or that the selected data region is too small?

[Shengfeng233]

你数值计算的结果可以通过升阻力系数，st数自行判定，关于你选择的区域大小问题，一般来讲不要太小就行（至少包含一些网格，不要在网格尺度以下），太小会出一些拟合问题。 具体你跑出来不收敛可能是其他的问题，你自己可以再检查一下，这个代码虽然比较老了，但是正确性是有保证的。 祝好， 盛峰 At 2024-02-27 22:12:06, "HCwinwinwin" ***@***.***> wrote: image.png (view on web) Thanks！ Before inputting the data, I processed it according to the steps mentioned, and it seems like there shouldn't be any issues. image.png (view on web) The results from the simulation in Fluent seem to be normal.Is it possible that the grid size of 0.5mm in my vortex shedding region is not accurate enough, or that the selected data region is too small? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[HCwinwinwin]

好的，谢谢大佬！