Friction-Brake Rotor Corrosion

Brakes that complain when they're barely used

The paradox: the least-braked vehicles have the worst brakes.

Complaints describe a grinding or groaning noise on the first few stops, visible rust on the rotor faces, and a soft initial bite that firms up after braking a few times. At service, rotors are flagged for surface corrosion. The pattern is counter-intuitive — these are low-mileage-on-friction-brakes vehicles — which is the first hint that disuse, not wear, is in play.

Because trim, driving mode, region, and rotor lot all move together, the symptom is operationally entangled. Usage telematics — brake-apply frequency, regen fraction, distance since the last firm stop — make a disciplined attribution possible.

The logged fleet

500 vehicles were logged over a service interval, pairing a rotor corrosion index (from service inspection / brake-noise telemetry) with usage signals and vehicle genealogy.

The dataset is synthetic, generated from a usage-and-corrosion model with a known ground truth so the method can be graded. Brake-apply frequency is the true driver; the rotor lot is built as a trim-fitment proxy with no metallurgical path of its own.

The field of suspects

H1 and H5 are the behavioral chain (a one-pedal default produces low apply frequency); H2 is an accelerant; H3/H4 are the tempting material explanations; H6 is the conventional wear assumption — which the data will turn on its head.

Where a hasty investigation ends

A Pareto ranks association. The continuous usage signals, held against one another, tell a different story.

What actually tracks the corrosion

The behavioral signals dominate, but lot, trim, and humidity all correlate. The four-gate test — association, significance, materiality, mechanism — separates the cause from the proxies.

A trim-fitment artifact

Lot Lr's correlation has a fitment explanation: Lr was installed mostly on Efficiency-trim vehicles, whose one-pedal default yields the lowest brake-apply rates. It labels "this car is driven one-pedal," nothing more. Two tests confirm it.

Second, the mechanism gate: rotor metallurgy and hardness for lot Lr are within material spec — there is no pathway by which the lot itself drives surface rust. Strong correlation, no mechanism: a confounder. H3 is rejected, and the supplier hold is released.

The brakes rust because they're never used

The regression makes the structure explicit. Lot explains some variance; humidity adds a little; adding the apply-frequency deficit dominates and drives the lot coefficient to near zero:

This is the abductive step, and it inverts the usual wear intuition. The best explanation for "the least-braked cars rust the most, lot just rides along" is that corrosion is driven by disuse: when the friction brakes are rarely applied, the rotor face is never mechanically scrubbed, so a rust layer forms and persists (until a few firm stops clean it off). Humidity sets how fast it forms; the one-pedal usage pattern sets whether it ever gets removed. The refined, testable claim — corrosion grows with distance since the last firm stop and resets when the rotor is scrubbed — is confirmed in §8.

Where the data meets the rotor

8.1 — Corrosion vs brake-apply frequency

If disuse is the cause, corrosion should be low for vehicles that brake normally and climb sharply below a usage threshold — with humidity shifting the whole curve. It does:

8.2 — The disuse signature

Tracking a single Efficiency-trim vehicle, corrosion accumulates with distance since the last firm stop and snaps back whenever a hard brake event scrubs the rotor — a sawtooth, the fingerprint of disuse:

8.3 — Behavior, mapped

The triage, distilled

The recurring per-vehicle decision compresses to three checks, evaluable from usage telematics plus the corrosion/noise signal. Each leaf names cause and action; thresholds come from the data (flag above corrosion ; low usage below ~1.5 applies/100 km).

Scrubbing the rotors without the driver

The mechanism chain and its confounder, side by side:

The fix restores the scrubbing the driver no longer provides, in software:

Forecasting susceptibility and field exposure

Because the cause is a quantifiable usage signal, both the per-vehicle susceptibility and the field exposure can be forecast. The susceptibility curve maps brake-apply frequency directly to flag risk:

Over a year, the field exposure for a low-apply (one-pedal) vehicle — and the effect of auto-scrub — is decisive:

Field & calibration action

Immediate: deploy the automatic brake-scrub routine fleet-wide via OTA, gated on distance-since-firm-stop and local humidity, with an imperceptible apply pressure. Corrective: tune the regen-blending default to raise baseline friction-brake usage slightly, and add humidity-aware scrub cadence for coastal/wet climates. Add driver guidance encouraging an occasional firm stop. Release the supplier hold on rotor lot Lr — it was never causal. Verify by confirming the corrosion sawtooth no longer crosses the threshold on low-apply validation vehicles, and track flag rate by apply-frequency cohort rather than by lot.

Kept or dropped

Reproducibility

The dataset and full analysis ship alongside as brake_corrosion_rca.py (usage-and-corrosion generator + diagnostic pipeline) and brake_fleet.csv (500-vehicle mockup). The core of the generator:

# rust accumulates when the friction brakes are underused
deficit  = max(APPLY0 - apply_freq, 0)              # APPLY0 = 3 /100km
corrosion = 5 + 16*deficit**1.3*(0.5 + humidity_norm) + 0.5*lotLr + noise
flag = corrosion > 35                                # service threshold

# confounder: rotor lot Lr fitted mostly to the Efficiency trim
p_Lr = where(trim_eff==1, 0.78, 0.20)               # proxy, metallurgy in spec

The pipeline reproduces every figure: the Pareto, correlation screen, partial-correlation confounder test (lot → r ≈ ), staged regression (R² ), the usage knee, the disuse sawtooth, and the prognosis.