What Happens Without Data Cleaning?
The filtered workflow removes samples where the reference floor is almost silent, the global response is below instrumentation noise, or the instantaneous shape diverges from the reference mode by more than 18°. The plots below show the behavior obtained before applying those checks.
Observed Issues Without Cleaning
- Large clusters appear at the opposite sign because low-energy segments flip phase randomly.
- The coefficient of variation of upper floors exceeds 40%, masking the physical hierarchy of floors.
- Gaussian envelopes stretch across the entire horizontal axis, suggesting non-existent modal coupling.
In short, the unfiltered analysis suggests that the first mode has comparable participation on all floors and even intermittent inversions—behavior that contradicts the measured story-drifts and the known physics of a shear-building. Filtering restores the monotonic profile and suppresses the spurious outliers.
Unfiltered Mode Shape Analyses
Each panel below is identical in layout to the main Step 1 plots, but uses the raw instantaneous shapes (no amplitude or shape-consistency filters).
Mode 1 — Unfiltered
Mode 2 — Unfiltered
Mode 3 — Unfiltered
How to Reproduce
- Open
mode_shape_analysis.py. - Call
compute_mode_shape_statistics(..., use_filter=False). - Regenerate
mode_#_analysis.htmlto see the raw scatter and CV values.
Switching back to use_filter=True enables the amplitude + shape consistency gates used throughout the report.
