So far, this whole diagnosis series has lived at an analyst’s desk — logs, a workstation, the full history of a car or a fleet, and time to think. But the very first diagnostician is the vehicle itself. Long before charging code P0AE0 reaches a fleet database, an electronic control unit decided to set it: in a few milliseconds, with a few kilobytes of RAM, with no labels and no second chances, under a hard rule that it must not cry wolf. This post drops down to that edge and builds the on-board detector from the ground up, then connects it back to the offline methods of the rest of the series.

It is the on-vehicle complement to Fault Detection and Isolation, which laid out the theory of residuals and isolation. Here we put that theory under the constraints of real embedded hardware — and we follow one fault all the way through: a high-resistance charge connector that sets P0AE0, caught by the car itself and handed up to the fleet pipeline.

What it covers

Ten sections, about twenty-two minutes, in plain language with diagrams and worked logic.

Part A · How a vehicle detects its own faults

§ 2 — Model-based observers & residuals. A model predicts a signal from the others; the residual is measured − predicted; threshold it to flag a fault. Parity relations and the Kalman/Luenberger observer in plain words, with a residual-monitor algorithm.

§ 3 — Limit, plausibility & rate checks. The cheap first line — range, stuck-at, and impossible-rate checks — running every loop beneath the model.

§ 4 — Edge ML. Lightweight on-vehicle anomaly detection: a tiny autoencoder trained on healthy data, and the four walls it lives in — memory, latency, no labels, drift (and OTA updates).

Part B · The constraints that shape it

§ 5 — The false-alarm budget & debouncing. Why on-board must be conservative, the real cost of a false MIL, and fault-maturation counters that store a code only after it trips N times under set-conditions.

§ 6 — Safety & certification. ISO 26262 / ASIL, determinism, freedom from interference, and auditability — why edge diagnosis is bounded where cloud ML is free.

Part C · Edge meets cloud

§ 7 — The closed loop. On-board detection emits a DTC, freeze-frame, and snapshot up to the fleet pipeline; fleet RCA tunes the thresholds and ships them back by OTA — the detector improves itself.

§ 8 — Worked example. Detecting the high-resistance connector on-board: an observer on charge voltage, a residual that grows under high current, maturation over several charges, then P0AE0 stored with a freeze-frame that seeds the fleet de-mix.

§ 9 — Practical notes. Field rules for diagnosis at the edge.

§ 10 — References & further reading. Fault diagnosis, estimation, functional safety, and embedded ML, with pointers.

Read it

Open the monograph →

The monograph lives at its own URL in the warm-paper layout of the series, with an observer/residual block diagram, an edge-ML pipeline figure, a debounce/maturation counter trace, the edge↔cloud closed-loop diagram, two algorithms, a worked P0AE0 example in verdict cards, practical notes, and references.

Where it sits

The diagnosis series, in order:

I · The core method

  1. Root-Cause Analysis Under Uncertain, Heterogeneous Evidence
  2. When Is the Evidence Enough?
  3. Harder Cases in Diagnosis

II · At fleet scale

  1. Fleet-Scale Root-Cause Analysis
  2. Fleet RCA, Made Rigorous

III · Proof, foundations & prevention

  1. Proving Cause with Experiments
  2. Diagnosis on the Vehicle — (this post)
  3. The Data Plumbing Behind Fleet Diagnosis
  4. From Root Cause to No Cause

Full map: The Diagnosis Series index.

← Back to Autonomy