Engineering / InfrastructureMedium (months)Detectability: Hard

Static corrosion growth rates in CP models

A cathodic protection (CP) integrity model used a single corrosion growth rate calibrated from the last campaign.

“Most failures begin as outdated confidence.”

Back to archiveDon’t end up here

Decision summary

Year
2019
Failure mode
Model inertia: a parameter assumed constant became the anchor that prevented re-seeing the system.
Silent failure window
~9 months: the model remained confidently stable while local conditions diverged.

The original logic

The last inspection campaign showed consistent wall loss behavior and the model matched field potentials and coupon readings; “locking” the rate reduced model churn and improved comparability across assets.

Key assumptions

  • The calibrated corrosion rate remained representative until the next campaign.
    Confidence at decision: High
    Expected lifetime: 12 months
  • Coupon readings and periodic potential surveys were sufficient proxies for corrosion acceleration.
    Confidence at decision: Medium
    Expected lifetime: 6–12 months

What changed

A coating degradation pattern shifted due to a change in seabed interaction (storm scouring and exposure). The CP system still “looked healthy” on averaged potentials, but local shielding and disbondment created pockets where the effective protection was worse.

Outcome

Unexpected defect growth in a short, previously benign segment forced rework of the CP model, increased inspection scope, and conservative operating limits during remediation.

Early warning signals (missed)

  • Growing spread (variance) in potential survey readings rather than the mean value
  • Repeat anomalies in ILI feature clustering that were attributed to sizing noise
  • Maintenance logs showing recurring localized coating repairs in the same corridor

How AssureAI would have helped

  • Track “rate stability” as an assumption with explicit expiry and evidence requirements.
  • Treat variance and clustering as first-class signals, not just headline averages.
  • Automatically link maintenance events and survey anomalies back to the model decision record.

Non-obvious lessons

  • A model that “doesn’t change” is not the same as a system that doesn’t change.
  • Averages hide localized risk; dispersion is often the early warning.
  • When you lock parameters, you should lock the validation cadence too.
Static corrosion growth rates in CP models — Decision Graveyard