Simulation from external input

[nbasse]

Hi!
I'm new to HOS-ocean (and wave theory, including HOS,) but we really like to use this powerful tool.

I and my colleague are currently writing our master thesis at the Mechatronics department, at the Royal Institute of Technology, Stockholm, Sweden.
We are researching wave prediction systems and their use for optimizing the performance of Wave Energy Converters (WEC), (and in particular systems based on Submerged Pressure Sensors (SPS) and HOSM)

We would like to use HOS-ocean to predict the development of a measured wave field (expressed in terms of eta and phis) through time and space. (Limited to 2D waves for now)
When going through the documentation and the code, I found input case (i_case) 32 that seems to be the type of function we need.

But when running a few tests, I found what I believe are some mistakes/bugs in the corresponding code.

First is regarding the reading of the input file (3d_ini.dat) that I managed to fix and have documented in this pull request:
568f52e

After doing the above changes I manage to start simulations based on a custom input file.
But, from looking at the code and doing tests with some regular waves, I have noticed that the simulation takes the parameter Tp from input_HOS.dat (for the JONSWAP) and uses it to calculate the length in x-direction how the waves propagate. (I.e if the Tp in input_HOS.dat doesn't match the initial wave field, the simulations looks wrong)
Is this the intended behaviour for case 32?
How does the inputted Tp relate to the calculations for the simulation of the wave field?

Is case 32 the best way of inputting a measured wave field as initial conditions?
Or can we use some other case?

Thank you for your time and sharing this software as opensource!
Best Regards, Niclas

[gducrozet]

@nbasse : Thank you for your interest in HOS-ocean.

After doing the above changes I manage to start simulations based on a custom input file.
But, from looking at the code and doing tests with some regular waves, I have noticed that the simulation takes the parameter Tp from input_HOS.dat (for the JONSWAP) and uses it to calculate the length in x-direction how the waves propagate. (I.e if the Tp in input_HOS.dat doesn't match the initial wave field, the simulations looks wrong)
Is this the intended behaviour for case 32?
How does the inputted Tp relate to the calculations for the simulation of the wave field?

This is indeed what is expected. In the suggested case, Tp is simply used to define the size of the domain, through the corresponding peak wave length. In the input file, xlen corresponds to the number of peak wavelength (and is consequently a non-dimensional quantity in the file). The same applies for the duration of the simulation, which is expressed as a number of peak periods in the input file.

Is case 32 the best way of inputting a measured wave field as initial conditions?
Or can we use some other case?

This is probably the simplest way. However, you will soon face one issue relative to specify the correct free surface velocity potential. When one knows the free surface elevation, it is not straightforward to obtain the free surface velocity potential in a non-linear context. A lot of works in the field of wave prediction try to solve this point thanks to different strategies.

Hope this helps,

Best regards,

Guillaume.

[nbasse]

@gducrozet Thank you for your answer.

This is indeed what is expected. In the suggested case, Tp is simply used to define the size of the domain, through the corresponding peak wave length. In the input file, xlen corresponds to the number of peak wavelength (and is consequently a non-dimensional quantity in the file). The same applies for the duration of the simulation, which is expressed as a number of peak periods in the input file.

After doing a lot of experiments with different settings and inputs, we have noticed that xlen seems to be involved in more calculations than just the ones for defining the non-dimensional and dimensional x-vector. For example, when comparing two identical simulations (one original from case 3 and one repeated from case 32) and setting xlen to a "wrong" value for the repeated run, the waves behave very differently. (Comparisons were done while looking at the elevation as a function of x, and x rescaled to match between the simulations).

Also, xlen seems to be rounded to an integer when used in the code, but has to be entered in the input file as a float.

Could be please explain how xlen affects the simulations and why?
So we can better understand how to set that variable when using HOS-ocean with an external input of unknown xlen and an estimated Tp

This is probably the simplest way. However, you will soon face one issue relative to specify the correct free surface velocity potential. When one knows the free surface elevation, it is not straightforward to obtain the free surface velocity potential in a non-linear context. A lot of works in the field of wave prediction try to solve this point thanks to different strategies.

Yeah, we have noticed that it is not as straight forward as we thought.
Now we use a linear formula to calculate the velocity potential for our current tests. But hope to expand it to handle nonlinear waves as well.
If you know some good articles or implementations on the matter of calculating it in a non-linear context, we would really appreciate if you could share them with us.
We have had a hard time finding any good info on this.

Thanks!

[gducrozet]

@nbasse

Could be please explain how xlen affects the simulations and why?

As said in my previous answer, xlen changes the length of the domain and is provided as a number of peak wavelengths. It induces also some changes regarding the discretization (i.e. with respect to the number of points in x-direction specified in variables_3D.f90). Then, what need to be setup is a combination of xlen and n1. The usual choice of the discretization is to be able to solve the larger frequencies in the spectrum. You can use the following rule : k_max = n1*2*pi/xlen = 5 k_p with k_max the maximum wavenumber solved and k_p the wavenumber at the peak of the spectrum. The value 5 is indicative, it can be pushed to 8 if you want an increased efficiency, with the possible consequence on stability detailed in: https://doi.org/10.1016/j.oceaneng.2017.07.003

Now we use a linear formula to calculate the velocity potential for our current tests. But hope to expand it to handle nonlinear waves as well.
If you know some good articles or implementations on the matter of calculating it in a non-linear context, we would really appreciate if you could share them with us.
We have had a hard time finding any good info on this.

There is a lot of literature that includes HOS in the prediction process. You can refer to the following work conducted in ECN some time ago: https://doi.org/10.1016/j.oceaneng.2010.03.002 or the following recent studies https://doi.org/10.1175/JTECH-D-19-0072.1 ;https://doi.org/10.3390/fluids5010009 ; etc.

[nbasse]

@gducrozet Thank you so much for your detailed response.

I haven't seen the article from Fujimoto and Waseda previously. Thank you for linking it!
We have read the rest of the linked articles before, but will definitely read them again and look more closely on the relevant parts.
Thank you.