Vibration Meter

A single-page web app for an Android phone mounted on the steering wheel. It records 3-axis acceleration, rotation rate and GPS speed. Every sample gets a timestamp. It then attributes the vibration to a probable cause.

No build step and no dependencies: plain HTML, CSS and ES modules. The app works offline once loaded, and you can install it to the home screen (PWA).

Running it

Browsers only expose motion sensors and GPS on HTTPS (or localhost).

• GitHub Pages: in the repository settings, go to Pages and deploy from the branch, root folder. Then open https://<user>.github.io/vibrationmeter/ in Chrome on the phone and choose Add to Home screen.
• Local test: python3 -m http.server 8000 and open http://localhost:8000. To reach it from the phone over HTTPS, use a tunnel (for example cloudflared or ngrok) or Chrome remote-debugging port forwarding.

Using it

1. Vehicle tab: enter the tire size (e.g. 205/55R16). Final drive ratio and cylinder count are optional.
2. Mount the phone rigidly on the wheel rim with the screen facing you, and press Start recording before driving. With the car parked and the wheels straight, tap Set steering centre. The screen is kept awake with the Wake Lock API. Android pauses sensors when the screen is off or the app is in the background.
3. Drive the pattern that produces the vibration. Include some standstill time with the engine running, steady speeds held for 20 s or more across the range, some firm stops and some strong accelerations. A passenger can tap Mark. For wheel bearings, drive long curves in both directions at 50–100 km/h. For CV joints, pull away at full lock both ways in an empty car park.
4. Recordings tab: analyze, export CSV (one row per sample) or JSON (raw streams, re-importable), or delete a recording. You can also import a CSV or JSON file, for example on a laptop, or load a synthetic demo drive.

Recordings are written to IndexedDB every 2 s while recording. A crash or closed tab loses at most a couple of seconds.

Data format (CSV)

column	meaning
timestamp_ms	epoch milliseconds (sub-ms resolution, from the sensor event time)
elapsed_s	seconds since the first sample
acc_x/y/z	acceleration including gravity, m/s², phone axes
lin_x/y/z	acceleration without gravity (if the phone provides it), m/s²
rot_alpha/beta/gamma	rotation rate, deg/s (alpha = around the screen normal = steering rotation)
speed_kmh	GPS speed interpolated to the sample time
lat, lon, gps_accuracy_m	last GPS fix
gps_age_ms	how old that fix is at this sample
steering_deg	estimated steering-wheel angle, degrees, + = left (can exceed ±180° at full lock)
lateral_acc	cornering acceleration, m/s², + = car accelerating to the left
longitudinal_acc	acceleration (+) / braking (−), m/s²
marker	1 where the Mark button was pressed, C where the steering centre was set
test	guided test running at that sample (engine_off, tap, idle, coast, …)

The last three are derived when exporting, so older recordings get them too.

The JSON export keeps the unmodified GPS fixes (1 Hz, with their own timestamps) next to the motion samples.

Guided tests

The Record tab has a guided-test list. A test records a labelled stretch under known conditions and starts the recording if needed. Parked tests are timed, with a countdown and beeps so your hands can stay off the wheel. Driving tests are started and ended by a passenger. You don't enter the speed where it vibrates: the analysis finds the worst speed band and judges the driving tests there. Tests that don't apply to the car are hidden, based on the Gearbox setting in Vehicle.

test	conditions	what it tells the analysis
Engine off	parked, wheels straight, hands off (15 s)	sensor noise floor, loose mount; sets the steering centre
Tap test	tap the rim once a second (10 s)	ringing frequency of steering column + mount; explains speed-band humps and fixed-frequency peaks
Idle	neutral, A/C off (20 s)	engine baseline; idle rpm from a comb over half/crank/firing orders; uneven running (misfire)
Idle with A/C on	(20 s, optional)	mounts / idle speed control / compressor, if vibration rises
In gear, brake held	automatics only (15 s, optional)	engine/gearbox mounts, if vibration rises well above idle
Rev steps	hold 1500 / 2000 / 2500 / 3000 rpm, 12 s each (optional)	vibration by rpm, resonance at one rpm, where each engine frequency lands after fold-back, and the neutral run-up verdict
Slow rev	to ~3000 rpm and back (30 s, optional)	resonances between the steps
Lock-to-lock	full left → full right (25 s, optional)	steering range; unequal left/right = wheel off-centre on the rack; shudder at lock
Speed steps	hold 30, 50, 70 … 130 km/h, ~15 s each	steady data over the whole range, to find the worst band
Coast in neutral	from the top speed down to ~40 km/h	at the worst band: unchanged → wheels/tires/brakes/bearings; disappears → engine or drivetrain under load
One gear lower	several speeds (optional)	changes → engine-rpm related
Pull away	manual only: from standstill in 1st, three times (optional)	clutch judder, flywheel, engine/gearbox mounts
Braking	firm, 90 → 20 km/h, twice (optional)	repeatable brake-judder data

Repeating tests. Every run is saved. Runs of the same test in a recording are pooled: their spectra are averaged, weighted by length. With three or more runs, one that is more than 40 % off the median is discarded. The table shows the run count and spread, and the app warns when repeated parked runs disagree. That usually means the radiator fan or A/C cycled, or the engine was still cold.

Engine frequencies and the 60 Hz limit. A 4-cylinder fires twice per crank turn. That is about 26 Hz at idle, which is measurable, but 50–100 Hz at 1500–3000 rpm, which is above the ~30 Hz limit. Those vibrations fold back to a lower frequency: at 2000 rpm, 67 Hz shows up at ~7 Hz. At a known rpm the fold-back is predictable, so the rev-step analysis labels such peaks as engine. A fixed-frequency peak while driving that sits on one of these folded engine lines is reported as probably engine-related.

Neutral run-up without knowing the driving rpm. If the parked engine stays well below the vibration of the worst driving speed band at every rev step, the engine is ruled out as the main source. If one step matches the driving level, the engine can explain it.

For the level-based comparisons (A/C, in gear), the app uses the larger of two ratios against idle: overall vibration, and vibration at the idle firing frequency. Coast and gear tests are compared with steady driving at the same speed (±6 km/h), within the worst band when the test reached it. Results appear in a Guided tests table in the Analysis tab and as findings. The CSV test column records which test each sample belongs to (rev_2000, coast, …).

Steering angle and cornering g

The phone turns with the wheel. Because the rim is tilted, gravity's direction across the screen rotates by the same angle: atan2(acc_x, acc_y).

• Gyro fusion. The gyro's rotation around the screen normal follows fast steering movements. Gravity removes the gyro's drift slowly. The result is unwrapped, so full lock (±400° and beyond) is tracked.
• Cornering g. Once the wheel angle is known, the part of the accelerometer reading across the rim is the car's sideways acceleration. It has no lag and is cross-checked against GPS (speed × heading change). GPS is the fallback when there is no gyro.
• Braking/acceleration g. This is read from the axis perpendicular to the screen. Its scale and sign are calibrated by regression against the GPS speed change.
• Zero. The zero is the Set steering centre tap. Without that tap, it is the average angle while driving straight, judged by a constant GPS heading.

How the analysis works

1. Motion samples are resampled onto a uniform grid. Gravity and steering movements are removed with a 0.4 s high-pass.
2. The drive is cut into ~4 s windows (50 % overlap). Each window gets a power spectrum, a mean GPS speed, a state (standstill, steady, accelerating, braking or varying, from braking/acceleration g) and a corner direction (left, straight or right, from cornering g).
3. Order tracking. Wheel rotation frequency is speed / tire circumference. Each window's spectrum is re-plotted against multiples of that frequency ("orders") and averaged. A vibration caused by the wheels lines up at the same order at every speed. Anything else smears out. Straight-line windows are used when there are enough of them.
4. Heuristics then look for these patterns:

pattern	probable cause
peak at 1× wheel order	wheel imbalance, bent rim, tire belt, hub/bearing run-out. Also seen in steering rotation → front wheels
peak at 2× wheel order	out-of-round / flat-spotted tire, radial force variation
peak at final-drive-ratio order	propshaft, U-joints, centre bearing
3× order only under acceleration	worn inner (tripod) CV joint
stronger only under acceleration	drivetrain: CV joints, engine/gearbox mounts
stronger only while braking (often 1× order)	brake disc thickness variation, sticking calipers
vibration at standstill	engine, mounts, misfire, A/C compressor
fixed frequency at all speeds	structural resonance, phone mount, or rpm-related
hump in one speed band	imbalance exciting a suspension/steering resonance
stronger in left corners only (or right only)	wheel bearing on the loaded side: left corners load the right wheels
stronger in all corners	suspension bushings, CV joints, cupped tires
stronger at > 90° steering while pulling away	outer CV joint
wheel held off-centre to drive straight (needs centre calibration)	alignment, tire pressure, sticking caliper, pulling tire
steering shake only on-centre (disappears in corners)	play in tie-rod ends or rack, plus imbalance

The charts show speed, vibration over time (braking and acceleration shaded, marks shown as ★), steering angle, vibration by corner direction, a spectrogram with 1×/2× wheel-order lines, a frequency-vs-speed (Campbell) map, the wheel-order spectrum, and vibration per speed band.

Limitations

• Android Chrome delivers devicemotion at about 60 Hz, so only frequencies up to about 30 Hz can be measured. That covers wheel orders (about 5–20 Hz at road speeds) and brake judder. It does not cover engine firing frequency while driving. Higher frequencies can alias.
• GPS speed updates about once per second and lags slightly. Order analysis therefore uses steady-speed windows primarily.
• Steering angle needs the wheel rim to be tilted (normal cars: 20–30°). Cornering g at walking pace is approximate. GPS cannot correct it at that speed, but the full-lock angle itself stays accurate.
• In a fixed gear, engine rpm is also proportional to road speed. To separate engine from wheel causes, repeat a run at the same speed in a different gear.
• Results are heuristic. Use them to guide inspection, not as a verdict.

Files

index.html            UI shell
css/style.css         styles (light and dark)
js/app.js             UI controller
js/recorder.js        devicemotion + geolocation capture, wake lock, chunked saving
js/storage.js         IndexedDB
js/format.js          data layout, CSV/JSON export and import, GPS↔sample alignment
js/analysis.js        FFT, order tracking, diagnosis heuristics
js/tests.js           guided test definitions and their analysis
js/steering.js        steering angle, cornering and braking g from gravity + gyro + GPS
js/charts.js          canvas charts with touch tooltips
js/sim.js             synthetic demo drive (baseline and rev steps, imbalance, brake judder, right wheel bearing, neutral coast)
sw.js                 offline cache