Flight Vibration Survey

Flight Vibration Survey (flight-test-operations/flutter/flight-vibration-survey)

Use when the task is reducing an in-flight mechanical vibration survey, rotorcraft main-rotor track-and-balance and airframe vibration limits and fixed-wing vibration or buzz surveys, from measured accelerometer time histories. This leaf implements the order-domain reduction in pure Python, stdlib only: per-rev (N/rev) order amplitudes from a synchronous DFT over integer-revolution windows, windowed total RMS survey levels, root-sum-square combination of the orders, and gating of each survey point against the declared vibration limit and the 1P trim limit.

Domain quick reference

• Integer-revolution window: N = round(m_revs * rate / rotor_hz), rounded to the nearest integer sample count; the window holds exactly m_revs rotor revolutions. The worked example uses N = 12 * 1000 / 5 = 2400 samples.
• Synchronous order DFT at order p: A_p = (2/N) |sum_{k=0}^{N-1} x_k exp(-j 2 pi (p m) k / N)| with m = m_revs. The p-per-rev component falls exactly on DFT bin p * m_revs of the m-rev window, so integer-revolution windows are exact, with no leakage.
• Total RMS over a segment: RMS = sqrt(mean(x^2)); this is the survey level of the window. windowed_rms slides with hop equal to the window and drops a trailing partial window.
• Root-sum-square of the orders: RMS_tot = sqrt(sum_p A_p^2 / 2), which equals the full-record RMS over integer revolutions for a pure multi-order signal.
• Vibration verdict: margin = (limit - level) / limit, pass when the margin is >= 0.
• 1P trim verdict: the same margin applied to the 1P amplitude against the trim limit; needs_trim when the amplitude exceeds the limit.
• Units: the accelerometer record is in g and the limits are declared in g; all reductions are unit-consistent.

Workflow

1. Fix the survey point: the accelerometer time history samples (g), the sample rate rate (Hz), the rotor frequency rotor_hz (Hz), the window length m_revs (integer revolutions, 12 in the worked example) and the declared vibration limit and 1P trim limit (g).
2. Extract each order of interest with order_amplitude(samples, sample_rate_hz, rotor_hz, order, m_revs). The synchronous DFT runs over the first N samples of the integer-rev window; a record shorter than one full window raises ValueError.
3. Get the survey level with total_rms over the window, or windowed_rms(samples, sample_rate_hz, window_s) when the record spans several windows and the per-window levels matter.
4. Combine the orders with rss_of_orders over the amplitude dict.
5. Gate the point with vibration_verdict(level_g, limit_g) and the 1P component with trim_verdict(amp_1p_g, limit_1p_g).
6. Reduce the whole point in one call with vibration_survey_summary(samples, sample_rate_hz, rotor_hz, orders, m_revs, vibration_limit_g, trim_limit_g); the returned dict carries the order amplitudes, the windowed total RMS, the RSS check and both verdicts. orders must include order 1 for the trim verdict.
7. Confirm the deterministic checks with the contract test.

Worked example

Rotor 5.0 Hz, 1000 Hz sampling, signal 0.15 g at 1P + 0.06 g at 2P + 0.08 g at 4P (deterministic sines with a phase offset per order), 12-rev window N = 2400 samples. Running the module gives these real outputs:

• Recovered order amplitudes: 0.150000 g at 1P, 0.060000 g at 2P and 0.080000 g at 4P, exact to 1e-6.
• total_rms = 0.127475 g and rss_of_orders = 0.127475 g (rounds to 0.12748 g); the RSS identity holds to floating-point round-off (difference about 5.6e-17).
• vibration_verdict(0.127475, 0.15): margin +0.150163, pass. The margin rounds to +0.150, the survey point clears the 0.15 g limit by 15 percent of the limit.
• trim_verdict(0.150000, 0.10): margin -0.500000, needs_trim. The 1P component sits 50 percent above the 0.10 g trim limit.
• vibration_survey_summary returns exactly the keys order_amplitudes_g, total_rms_g, rss_of_orders_g, vibration_verdict and trim_verdict.

Pitfalls

• Choosing a window that is not an integer number of rotor revolutions: the 12-rev window (N = 2400 at 1000 Hz) centers the order bins, and a non-bin-centered tone (2.3P) leaks up to the 0.05 g sidelobe envelope while a bin-centered off-order tone (2.5P) stays below 1e-6 g.
• Judging the survey by total RMS alone: total_rms equals the RSS of the orders (0.127475 g), but the 1P trim check runs against its own 0.10 g limit - 0.150000 g at 1P is needs_trim even though the total passes the 0.15 g vibration limit.
• Misreading the margin signs: +0.150 against the 0.15 g limit is a pass with 15 percent clearance, -0.500 against the 0.10 g trim limit means the 1P component sits 50 percent above, and zero margin at the boundary counts as pass or no-trim.
• Reading order amplitudes off the phase of the signal: the recovery is phase invariant (the fixture carries a phase offset per order), so amplitude estimates do not depend on when the window starts.
• Feeding non-physical inputs: empty samples, non-positive sample rate, rotor frequency, or window length, order below 1, m_revs below 1, non-positive vibration or trim limits, negative levels, and a record shorter than one full window all raise ValueError.
• Underestimating the record length needed: the reduction needs at least one full window of samples, so a survey point recorded shorter than the window is rejected rather than partially analyzed.

Verification

• Confirm order_amplitude recovers 0.150000 / 0.060000 / 0.080000 g to 1e-6 on the worked signal, and that recovery is phase invariant.
• Confirm rss_of_orders equals total_rms to 1e-6 (the identity holds over the 12-rev window).
• Confirm the verdict margins: +0.150 pass against the 0.15 g limit and -0.500 needs-trim against the 0.10 g trim limit, with zero margin at the boundary counting as pass or no-trim.
• Leakage: a 2.5P off-order tone stays below 1e-6 g in every integer-order bin (bin-centered orthogonality), and a non-bin- centered 2.3P tone stays below 0.05 g, the sidelobe envelope, in every order bin.
• Confirm every non-physical input raises ValueError: empty samples, non-positive sample rate, rotor frequency or window length, order below 1, m_revs below 1, non-positive vibration or trim limits, negative levels, and a record shorter than one full window.
• Run the contract test offline: python3 scripts/test_flight_vibration_survey.py (35 tests, deterministic, under 20 s).

Related leaves

• flight-test-operations/flutter/ground-vibration-testing: the ground-based sibling; this leaf covers the in-flight survey.
• flight-test-operations/performance/rotorcraft-performance-flight-test: rotorcraft performance testing that sits alongside the rotor balance survey flight work.
• flight-test-operations/envelope/buffet-boundary-testing: the high-Mach envelope sibling; this leaf reduces the steady in-flight survey points.

Behavior contract (gate 3)

Run the deterministic contract test (stdlib unittest, offline):

python3 scripts/test_flight_vibration_survey.py

The test covers exact 1P/2P/4P amplitude recovery to 1e-6 (single tone, multi-order and phase-shifted), the total-RMS equals RSS-of-orders identity at 0.12748 g, windowed RMS over one, two and partial windows, the verdict margins of the worked example, the documented survey summary key set, leakage rejection of 2.5P and non-bin-centered off-order tones, the integer-rev window rounding convention, run-to-run determinism, and ValueError rejection of non-physical inputs.

Compliance

• Standards referenced, not reproduced: FAR 29 for the rotorcraft track-and-balance and vibration survey airworthiness context, and the FAR 25 / CS 25 vibration and buffeting context of 25.251 for fixed-wing survey work; name-and-paraphrase references only per standards-map.yaml.
• compliance: STANDARDS-REF, gated: false.