Frame size for time domain input data

Can the frame size ( number of samples) be set by the user and what is the minimum frame size for feature extraction processing ( Spectra , MEL …)

Hi @pjacquesd,

The frame size can be set through the window size in the Impulse Design. For instance with data sampled at 100 Hz, a one-second window would contain 100 samples.

Regarding the feature extraction, the frame length can be set to 1 ms or even less. However this will increase the extraction time on target (you can check the On-device performances when changing those values).
You can also get more information on processing blocks here: https://docs.edgeimpulse.com/docs/processing-blocks

Aurelien

I have vibration data in CSV files : first column with header “time stamped” and data per sample interval of 1/2e3 = 0.0005 seconds and a total of 2000 samples

image

I have 10 of these files with normal vibration data and 10 with faulty bearing vibration.
I want to extract spectral or flat features and do machine learing

Hello @pjacquesd,

Quick note on your timestamp.
timestamp values should be millisecond in our CSV ingestion format. Here you have seconds.

Regards,

image

Hi @pjacquesd,

I just had a look at your project, it seems you solved your issue. Did it come from the timestamp in seconds instead of milliseconds?

Regards,

Louis

@pjacquesd An addition to @louis when you set the impulse length to e.g. 1000ms. and your data is at 1000Hz you’ll need 1000 samples in the file. If there’s not enough (and zero padding is disabled) you’ll get this error. So probably your frequency is wrong, and that’s why it’ll yield to this.