Step 1: Mode Shape Extraction

Extract and normalize mode shapes from acceleration data using RMS averaging

Mode Shape Matrix

The extracted mode shapes are summarized in the mode shape matrix $\boldsymbol{\Phi}$, where each column represents a mode:

\[ \boldsymbol{\Phi} = \begin{bmatrix} 0.3089 & -0.9694 & 1.0000 \\ 0.7034 & -0.6746 & -0.9225 \\ 1.0000 & 1.0000 & 0.4605 \\ \end{bmatrix} \]

where $\phi_{i,j}$ represents the normalized mode shape component at floor $i$ for mode $j$. The matrix is normalized such that the maximum absolute value in each column is 1.0. The actual numerical values are displayed in the individual mode analyses below.

Methodology

The workflow assumes a single dominant mode during each forced-vibration record:

  1. RMS ratios: floor accelerations are converted to relative modal amplitudes via \( \phi_j/\phi_3 \approx a_j/a_3 \).
  2. Sign consistency: correlations with the roof response enforce in- / out-of-phase behavior.
  3. Instantaneous shapes: every time sample produces a normalized acceleration vector aligned with the reference shape.
  4. Cleaning filters: low-amplitude or off-shape samples are discarded (reference floor <5% of max, global response <2%, or cosine <0.95).
  5. Statistics: the remaining samples feed the Gaussian and time-variation plots; raw (unfiltered) moments are logged for traceability.

Mode 1 Analysis

Mode 2 Analysis

Mode 3 Analysis

Statistical Analysis Summary

Key Metrics

The visualizations above show four key analyses for each mode:

1. Normalized Mode Shape
Reference shape from RMS ratios, normalized to max amplitude of 1.0 and aligned with the roof motion.
2. Coefficient of Variation
Computed both for the raw scatter and the filtered set; reported CVs correspond to the used (filtered) samples.
3. 95% Confidence Intervals
Bounds of the filtered instantaneous shapes. Raw limits are logged in the console for comparison.
4. Time Window Variation
Grouped averages of the retained samples (1-second bins) contrasted against the reference modal amplitudes.

What If We Skip the Filters?

The cleaning stage is optional but strongly recommended. Without removing low-energy and off-shape samples, coefficients of variation can exceed 40%, Gaussians span both signs, and the time-variation plot becomes dominated by noise. A dedicated summary of the raw, unfiltered analysis is available below.

View Unfiltered Sensitivity
Back to Menu View Combined Analysis