The London Eye: A Sound & Vibration Field Study from Inside a Capsule
Revised 5 July 2026 to correct the interpretation of the low-frequency harmonic series.
Riding the London Eye on 5 July 2026, I ran an impromptu two-sensor experiment from my seat: a smartphone accelerometer record (~400 samples/sec, 50 seconds) and an audio recording (44.1 kHz, 40 seconds). The question: what does a 135-meter observation wheel actually look like in the frequency domain — and do the two sensors tell the same story?
They do not. And that turned out to be the most interesting result.
The Wheel as a Rotating Machine
The London Eye is, structurally, an enormous bicycle wheel turned on its side and stood upright: a cable-stayed rim, 120 meters in diameter, supported by tensioned spoke cables radiating from a central hub, with 32 passenger capsules mounted outboard of the rim. The rim is driven by motorized tires at the base rather than through the hub — the hub merely holds tension.
The wheel’s rotation rate is stately. The rim moves at roughly 26 cm/s, completing one revolution in about 30 minutes. That corresponds to a rotation frequency near 0.00056 Hz — a fundamental so low that no practical field measurement resolves it directly as a spectral line. Everything interesting therefore happens at frequencies set by faster-turning elements in contact with the rim, or in the local machinery riding along with you.
From a machinery-diagnostics standpoint, this makes the Eye an unusual specimen: a rotating machine whose shaft speed is four orders of magnitude below the tones you actually measure.
Instrumentation
Two channels, one ride, same capsule:
- Triaxial accelerometer (smartphone MEMS, ~400 samples/sec, 50-second record) — capturing structure-borne vibration transmitted through the capsule mounting.
- Microphone (44.1 kHz, 40-second record) — capturing the acoustic environment inside the capsule.
Both records were processed with standard FFT and power spectral density methods, with peak identification against candidate sources derived from the wheel’s geometry and the capsule’s equipment.
Vibration Results (Accelerometer FFT)
| Frequency | Identified Source |
|---|---|
| DC drift, Y-axis (38.5° over 50 s) | Wheel rotation — gravity vector turning through the capsule frame |
| 0.88 Hz | Discrete low-frequency tone — consistent with a small roller in rolling contact with the rim (see arithmetic check below) |
| 1.17 Hz | Capsule pivot/sway mode |
| 2.0–3.4 Hz | Harmonics 3x–5x of a ~0.67 Hz base rate |
| 4.0–4.5 Hz | Related low-frequency harmonic content |
| 6.0–7.4 Hz | Harmonics 9x–11x of the same ~0.67 Hz base — a classic harmonic series |
The harmonic series dominated the radial (Z) axis. Even spacing at ~0.67 Hz across orders 3x through 11x is the fingerprint of a true harmonic series rather than a cluster of unrelated structural modes — the same reasoning used to separate gear-mesh families from resonances in gearbox spectra. The question is what rotates at 0.67 Hz.
Checking the Arithmetic: Why It Is Not Spoke or Gondola Passing
My first instinct was to attribute the series to wheel geometry — spoke-passing and gondola-passing frequencies, by analogy with blade-passing in a fan. The arithmetic does not support it.
A passing frequency is the number of repeating features multiplied by the rotation rate:
- Gondola passing: 32 capsules × 0.00056 Hz ≈ 0.018 Hz
- Spoke passing: 80 spoke cables × 0.00056 Hz ≈ 0.045 Hz
Both are far below the measured tones — the 0.88 Hz peak is roughly 49× the gondola-passing rate, and the 0.67 Hz base rate is roughly 15× the spoke-passing rate. The wheel simply turns too slowly for its own geometry to generate spectral lines in the measured band. The 30-minute revolution period rules the geometric interpretation out.
So the excitation must come from something turning much faster than the wheel itself: a rotating element in rolling contact with the rim. The kinematics are simple. A roller of diameter d in contact with a rim moving at velocity v rotates at
f = v / (πd)
At the measured rim speed of 26 cm/s:
- A roller of ~12 cm diameter turns at ~0.67 Hz — matching the harmonic series base rate
- A roller of ~9 cm diameter turns at ~0.88 Hz — matching the discrete tone
These are plausible sizes for guide, restraint, or support rollers along the rim, or for elements within the drive-tire stations at the base. A slightly out-of-round or unbalanced roller in continuous contact with the rim produces exactly this signature: a once-per-revolution forcing at its own rotation rate, rich in harmonics, transmitted into the structure. I have not yet confirmed the specific hardware, so this remains a hypothesis — but it is the one the kinematics support. The alternative for isolated tones is low-order structural modes of the capsule mounting, which is the assignment retained for the 1.17 Hz sway peak.
The slow Y-axis gravity drift is a bonus signature: over the 50-second record the sensed gravity vector rotated through the capsule frame, meaning the accelerometer doubles as an inclinometer tracking the capsule’s arc around the wheel.
Overall vibration levels were low — σ < 1.3 m/s² including the gravity drift. The ride is genuinely smooth. The structure-borne signature is dominated by rolling-contact elements and capsule dynamics, not by the drive machinery.
Sound Results (Audio FFT)
| Frequency | Identified Source |
|---|---|
| 49.1 Hz + clean harmonics through 6x | Capsule AC compressor/fan — 2-pole induction motor near 2946 RPM, slip just below the 50 Hz UK supply |
| 100 Hz + harmonics | Mains hum (2× 50 Hz) from power electronics and lighting |
| 630–640 Hz cluster | Probable AC fan blade-passing frequency (49.1 × 13 = 638 Hz) |
| 300–3400 Hz bursts | Passengers’ voices — the classic telephony band |
The 49.1 Hz tone is a textbook induction-motor signature. A 2-pole machine on a 50 Hz supply has a synchronous speed of 3000 RPM; the measured 49.1 Hz fundamental corresponds to about 2946 RPM, or roughly 1.8% slip under load. Its harmonics ran cleanly out to 6x and held steady across the entire record, consistent with a compressor or fan running at constant duty.
The 630–640 Hz cluster fits a blade-passing interpretation: 49.1 Hz shaft speed multiplied by a 13-blade fan gives 638 Hz. Note that here the passing-frequency arithmetic works, because the fan shaft actually turns at ~49 Hz — unlike the wheel. The microphone thus identified the motor type, pole count, approximate load condition, and a plausible blade count — without ever seeing the nameplate. This is the same envelope of inference used every day in machinery condition monitoring, applied here to a tourist attraction’s air conditioner.
The Key Takeaway: Two Sensors, Two Different Machines
Here is the punchline. From the same seat, at the same time:
- The accelerometer saw the wheel-side mechanics — a coherent harmonic family most plausibly from rollers in rolling contact with the rim, the capsule sway mode, and the rotation itself via gravity drift.
- The microphone heard the capsule’s own local machinery — the air-conditioning system — and its occupants.
The main drive motors, more than 60 meters away at the base of the wheel, were essentially invisible in both records. The long transmission path through the tensioned rim and spoke cables attenuates their structure-borne signature, and inside the capsule the local AC system dominates the acoustic floor.
The broader lesson for anyone doing machinery diagnostics or field measurements: your sensor choice determines which physics you observe — and your first source hypothesis deserves an arithmetic check before it goes in the report. A passing frequency that fails the N × frotation test by an order of magnitude is telling you to look for a different rotating element. The vibration and acoustic channels here were not redundant — they were complementary, each essentially blind to the other’s dominant source.
A Note on the Passive Pivot
One design detail worth appreciating: the London Eye’s capsules ride on a passive, gravity-stabilized pivot. There is no active leveling motor keeping the floor horizontal as the capsule travels around the rim — gravity does the job. Accordingly, no leveling-motor signature appears in either spectrum. Sometimes the absence of a peak is itself a finding, and here it confirms a design choice: fewer motors, fewer tones, fewer things to maintain 120 meters in the air.
Not a bad data set for the price of one admission ticket.