princetonuniversity / magnet Goto Github PK

As the results of: 1) the units of f, B and P have been changed, and 2) the definition of duty ratios for trapezoidal wave has been changed, the assertion values in the test_coreloss.py need to be modified.

To avoid mistakes, @ds9056 let's double check the codes and define the new assertion values for future tests.

Is there any plan to include other topologies such as an LLC, Weinberg or active clamping Flyback or Forward?

Thank you very much!

And my most sincere congratulations, you have done a great work!

Add B-H curve display session.

I was wondering which materials will be included in your MagNet tool soon. I am particularly interested in materials ML95S and ML91S from Hitachi. Will these ones be included at some point?

Thank you for your great work!

The following approach is incorrect and will fail as soon as data is moved around within the package context:

We should be using importlib_resources here.

Ideally, setup.cfg should list all developer requirements in the dev section (like streamlit) such that developers doing a pip install -e .[dev] will get all the dependencies they need. A requirements.txt can still be provided for a more fine-grained experience of course.

Modify Bmin/Bmax to be in T / change slider range

I'm trying some fitting approaches, but I have different results. Do you use the same method to calculate the Steinmetz coefficients for iGSE_trapezoid and iGSE_sine?

For exemple, for 3C94 material fitting datasheet data I have: [K, alpha, beta] =[3.82978222, 1.38462252, 2.73482858]
with: freq =100e3 and flux = 0.328.

So, for duty_ratios = [0.1, 0.1, 0.4], I have:
magnet.core.core_loss_iGSE_trapezoid(freq, flux, duty_ratios, K, alpha, beta, None, 0)
49101373.64068319

And defining the material I have:
magnet.core.core_loss_iGSE_trapezoid(freq, flux, duty_ratios, 0, 0, 0, '3C94', 0)
4065393.854135263

I want to know your method to understand this difference :)

Dear team,

I can see now that all the database, as well as the options you provide, only allows a frequency range from 50 to 500 kHz.

Nevertheless, many applications require frequencies over 500 kHz.

Are you planning to extend your tool for frequencies over 500 kHz (for those materials that can operate over this frequency) ?

Thank you very much!! Keep up the good work.

Testing 3C94 material with triangular waves under MagNet AI showed that losses are not symmetric in duty cycle. Why?

Would it be possible to add a button to remove one of the plots?

I am going to make changes that affect how the data is displayed. For Triangular or Trapezoidal waveforms, it is not easy to visualize losses when more than a duty cycle is displayed. I would suggest either depicting a single duty cycle (type of waveform) at a time. Or showing the plots for different duty cycles (types of waveforms) in separated plots one after the other. Any comments? How do you think it will look better?

I ran through a full training process of LSTM and got a trained LSTM model for N27. It works good on our dataset (see attached figure), but bad with user defined input from the website, even with the simplest triangle wave.

Still working on it to see if it's due to the mis-coding of the webpage or some issue from the model itself.

BTW: The iGSE results for Arbitrary section is now correct and consistent with what we may get from the Triangular and Trapezoidal sections.

The following comment/code suggests that either the excitations constant needs changing, or some data is no longer needed.

The comment about waveform consolidation also suggests we need to revisit the places where the old excitation are used. In any case, this entails changing code after careful consideration.

I'm wondering if this is correct:
https://github.com/PrincetonUniversity/Magnet/blob/af49eb1108079c2c2d072d1c3224d0ee1df2c09c/main.py#L323

Should the integral be evaluated from 0 to T, like in the other waveforms (example below):

def calc_iGSE_tagl(Freq, Flux, flux_list, duty_list, ki, alpha, beta):
    time = np.linspace(0, 1 / Freq, 10001)
    ...

Describe the bug
'Download Raw Data' generates an empty .csv file

To Reproduce
Steps to reproduce the behavior:

1. Go to 'https://mag-net.princeton.edu/'
2. Chose a material
3. Download Raw Data
4. The downloaded .csv-file is empty

Expected behavior
Downloaded .csv-file contains data

It's found that the units of data (mostly, the flux density T/mT and the core loss W/kW) are not consistent among the dataset, data visualization and core loss prediction. Currently doing a full check and uniforming them according to the [SI] system in the coming branch.

First of all, congratulations on the work done. I look forward to using the tool in my future research.

In the section about the calculation of losses, it is stated that the estimation of losses comes from the average of the instantaneous voltage and current multiplication. This would include both winding and core losses.

Are you neglecting them? Are you designing them to be neglectable? does this work at all operation points?

It would be good to clarify.

Best regards.

Added the FAQ section in the hrl-faq commit, which can be accessed from the sidebar. There are currently three questions there. Let me know if there's anything else to be included, or directly include them in ui_faq.py

Streamlit UI options like:
https://github.com/PrincetonUniversity/Magnet/blob/598b417a45583b6c42b02f06f3f96b605284a8ab/app/ui.py#L98

should inside magnet.ini's [streamlit] section:
https://github.com/PrincetonUniversity/Magnet/blob/598b417a45583b6c42b02f06f3f96b605284a8ab/src/magnet/magnet.ini#L4

but this will entail supporting lists of values.