Facebook
Google
GitHub
Linkedin
Hi everyone,
I’m currently working on processing experimental data from a sensor calibration setup. In each test, a force of 10 N, 20 N, and 30 N was applied sequentially to a structure via a tensioned cable, with about 10 seconds between each load step.
The setup records both the applied force and the corresponding bridge sensor signals. Due to mechanical tension in the setup, the cable always has a small preload, so the force signal never truly reaches zero.
After performing DTW-based alignment between the force and bridge signals, I apply a signal-to-noise ratio (SNR) filter to exclude noisy bridge channels before calibration.
Currently, I compute SNR as the RMS of the smoothed signal divided by the variance of the residuals (signal – Savitzky–Golay smoothing). Channels with SNR < 30 dB are discarded.
I’m wondering
1) is this way of estimating SNR and filtering signals scientifically valid, given that the residual also contains high-frequency signal components (not just noise)
-Are there better or more established methods for filtering or weighting bridge signals in such calibration experiments (e.g., coherence analysis, robust regression, etc.)?
Any feedback or references on best practices for noise estimation and peak filtering in sensor calibration data would be greatly appreciated.
Thanks in advance!
I’m currently working on processing experimental data from a sensor calibration setup. In each test, a force of 10 N, 20 N, and 30 N was applied sequentially to a structure via a tensioned cable, with about 10 seconds between each load step.
The setup records both the applied force and the corresponding bridge sensor signals. Due to mechanical tension in the setup, the cable always has a small preload, so the force signal never truly reaches zero.
After performing DTW-based alignment between the force and bridge signals, I apply a signal-to-noise ratio (SNR) filter to exclude noisy bridge channels before calibration.
Currently, I compute SNR as the RMS of the smoothed signal divided by the variance of the residuals (signal – Savitzky–Golay smoothing). Channels with SNR < 30 dB are discarded.
I’m wondering
1) is this way of estimating SNR and filtering signals scientifically valid, given that the residual also contains high-frequency signal components (not just noise)
-Are there better or more established methods for filtering or weighting bridge signals in such calibration experiments (e.g., coherence analysis, robust regression, etc.)?
Any feedback or references on best practices for noise estimation and peak filtering in sensor calibration data would be greatly appreciated.
Thanks in advance!
