Mechanism Attribution Accuracy (MAA)

GMI-PI-005Process IndustriesIntegration-RequiredPattern O-7: Inspection Coverage

Definition

Percentage of AI corrosion predictions where the assumed damage mechanism was confirmed by subsequent physical inspection: detecting whether the model is predicting the right type of damage, not just the right rate, since mechanism errors change remaining-life estimates by an order of magnitude.

The Measurement Gap

AI corrosion prediction reports a rate. Few models report which mechanism produced the rate, and fewer programs verify whether the assumed mechanism matches what the inspector actually finds. The failure mode is mechanism substitution: the model trained on flow-accelerated corrosion data and applies the same logic to a circuit where the dominant mechanism is stress corrosion cracking. The rate looks plausible. The remaining-life implication is wrong by an order of magnitude. Inspection adherence, model accuracy on rate, and corrosion rate trend all report the rate; none of them report the mechanism. Inspection findings reports list what was found; those findings rarely loop back to the model's assumption. The model can score "accurate" on rate prediction across a quarter while being systematically wrong on mechanism on every circuit where the dominant damage type has shifted.

Decision Supported:

Is the AI predicting corrosion from the right mechanism, or is it modeling flow-accelerated corrosion when the actual damage is stress corrosion cracking, a distinction that changes remaining-life estimates by an order of magnitude?

Formula

Plain language: Count the AI corrosion predictions that were followed by a physical inspection within the prediction's validity window. Among those, count how many had the AI's assumed mechanism confirmed by inspection findings. Divide and multiply by one hundred.

Notation:

MAA = ( Correct_mechanism_attributions / Total_AI_predictions_with_subsequent_inspection ) × 100

Variables:

Variable Definition Unit
Correct_mechanism_attributions Predictions where the AI's assumed mechanism was confirmed by inspection findings (visual, metallurgical, or failure analysis) count
Total_AI_predictions_with_subsequent_inspection Predictions followed by a physical inspection within the prediction's validity window count
Mechanism categories General corrosion, pitting, crevice, galvanic, flow-accelerated, erosion-corrosion, SCC, MIC, under-deposit, high-temperature oxidation, sulfidation, hydrogen attack, naphthenic acid n/a

Report MAA aggregate plus a confusion matrix showing which mechanisms the model confuses with which others.

Proposed formula, pending validation pilots.

At a Glance

Unit
Percentage
Trend Direction
Higher; Increasing toward 100%
Reporting Frequency
Quarterly, with quarterly confusion matrix
Category
Predictive Integrity and Corrosion Management
← Inspection Deferral Risk Accumulation (IDRA)Twin Calibration Drift Index (TCDI) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment