This is Case 3 of the five-part RCA series. After a software bug and an electromagnetic-structural interaction, this one is the manufacturing failure family — where the cause is on the production floor and the fix is a process-control change.

The complaint: elevated tab-to-busbar weld resistance in finished battery modules, creating thermal hotspots that fail acceptance testing. The Pareto chart, on first glance, points straight at one supplier lot of weld electrodes — a convincing story, well within the precision of the data. It is also wrong. The investigation has to refuse the easy answer and look at the actual process — the welding stations, the electrodes on those stations, and the maintenance interval at which the electrodes are dressed. The lesion is electrode-tip wear under a dressing interval that drifted longer than the SPC charts could catch with their current setpoint. The fix is a tightened dressing interval plus a tip-condition sensor — the kind of fix that costs almost nothing to deploy and dramatically improves yield.

This case demonstrates the manufacturing-RCA discipline that is its own subculture inside the broader RCA field. The vocabulary is different (Cp/Cpk, SPC, dressing intervals, tooling wear), but the method is the same: refuse the convincing impostor, walk the data back to the actual mechanism, validate the fix.

What it covers

Eleven sections, about eighteen minutes of careful reading.

§ 1–2 — The yield problem. The acceptance-test failures. The financial pressure. The first instinct that the lot is bad.

§ 3–4 — The Pareto and why it’s wrong. The supplier-lot chart that looks definitive. The cross-validation that breaks it. The discipline of distinguishing correlation in the Pareto from causation in the process.

§ 5–7 — Walking the process. The actual weld stations. The electrodes. The dressing intervals. The tip-wear curves. The data that surfaces when you stop staring at the supplier dimension and start staring at the manufacturing dimension.

§ 8–9 — The lesion: dressing-interval drift. The maintenance schedule that crept longer over six months. Why the SPC charts missed it. The control-chart redesign.

§ 10–11 — The validated SPC fix. Tightened dressing interval. Tip-condition sensor. The yield-curve recovery. The before-and-after Cpk numbers.

Read it

Open Case 3 →

The report lives at its own URL with a warm-paper manufacturing-document layout — steel-blue accent, classical SPC-chart styling throughout.

Code & data

↓ manufacturing-rca-code-and-data.zip — the Python generator (weld_resistance_rca.py), the synthetic 600-module dataset (weld_modules.csv), and a README.

The full series

Case 3 of 5. Read the rest in order:

  1. Battery Degradation Case Study — software / firmware
  2. Preconditioning-Mode NVH — electromagnetic / structural
  3. Module Weld Resistance — (this post) — tooling / manufacturing
  4. HV Isolation Faults — environmental (condensation)
  5. Brake Rotor Corrosion — usage / behavior

Followed by the agentic framework and its worked walkthrough.

← Back to Autonomy