ILI Threat Mechanism Coverage Index (ITMCI)

GMI-OG-014Oil and GasIntegration-Required

Definition

ILI Threat Mechanism Coverage Index, or ITMCI, is the ratio of threat mechanisms with adequate AI detection capability to the total threat mechanisms applicable to a pipeline segment under the operator''s PHMSA-required threat assessment.

The Measurement Gap

Patricia is the integrity engineering manager for a midstream operator running about thirty-eight hundred miles of liquids pipeline across three states. Two years ago her company contracted with an ILI service provider whose anomaly detection pipeline incorporates an AI scoring model that ranks indications by severity and likelihood of being a true defect. The vendor''s pitch was the obvious one: faster turnaround on tool runs, fewer marginal indications elevated to digs, dig hit rates up from sixty percent to almost eighty. The first year of operation under the new scoring delivered on those numbers. Dig efficiency improved, integrity program costs came down, and PHMSA reportable events stayed flat.

The wake-up arrived in the second year. A segment in the southern part of the system suffered a leak that should not have happened. The post-incident dig revealed selective seam weld corrosion in a pre-1970 ERW pipe section, with a defect that had grown to about sixty-eight percent of wall over what looked like the last three years. The most recent ILI run had been eighteen months earlier. The AI had identified anomalies in the segment but had not elevated this particular feature; the scoring assigned it a low priority based on its geometric signature, which did not match the patterns the model treated as high-severity. The defect was real. The tool had captured it. The AI had not flagged it.

When Patricia''s team pulled the AI scoring model''s validation history, the pattern became clear. The vendor''s training data came from over six thousand verified dig reports across multiple operators. The data was dominated by general corrosion and external pitting because those are by far the most common mechanisms in the dig record. Selective seam weld corrosion was represented in the training data by only forty-one verified examples. Stress corrosion cracking was represented by sixty-eight. Hard spots were represented by twelve. The model''s overall validated recall was eighty-three percent (a number the vendor reported prominently) but the recall was a population-weighted average that smoothed over the under-represented mechanisms. The recall on selective seam weld corrosion specifically, in the small validation slice available, was below forty percent. That number was not on any vendor slide.

The integrity program had been operating under a PHMSA-required threat assessment that listed selective seam weld corrosion as an applicable threat on pre-1970 ERW segments. The assessment was correct. The AI scoring''s coverage of that threat was inadequate. The two facts had not been connected in writing, and the integrity program had been quietly relying on an AI model that was strong on common threats and weak on the threats most likely to cause a serious incident.

Legacy ILI program metrics do not catch this. Dig hit rate measures whether the digs the operator chose to do were productive: it is silent on the digs the operator did not do because the AI did not flag them. Anomaly count by severity reports what the AI scored, not what the AI missed. Run-to-run trending reports growth rates on features the AI saw twice; it does not report features the AI saw the first time and underscored, then missed the second time when they had grown into a higher severity band.

What is missing is a metric that holds the AI''s threat coverage against the threat assessment that PHMSA requires the operator to defend.

Formula

ILI Threat Mechanism Coverage Index, or ITMCI, is the ratio of threat mechanisms with adequate AI detection capability to the total threat mechanisms applicable to a pipeline segment under the operator''s PHMSA-required threat assessment. The KPI exists to answer one decision: does the AI ILI scoring program provide complete threat coverage as the operator''s integrity management plan claims, or is the program optimizing for common mechanisms while underscoring rare but critical ones the threat assessment requires the operator to address?

The structural form is a count divided by a count. For each pipeline segment, list the applicable threat mechanisms from the segment''s threat assessment: general corrosion, pitting, selective seam weld corrosion, stress corrosion cracking, mechanical damage, manufacturing defects, hard spots, and whatever else the assessment names. For each mechanism, determine whether the AI scoring model has been validated to detect anomalies of that type at a recall rate above a minimum threshold. ITMCI is the count of covered mechanisms divided by the count of applicable mechanisms.

The minimum recall threshold is the operative judgment in the metric. Proposed starting values, with explicit operator recalibration as evidence accumulates: a recall of 0.70 for non-time-dependent threats and a recall of 0.85 for time-dependent threats. Time-dependent threats (corrosion, cracking, fatigue) are where missed detections compound into serious incidents over the inter-inspection interval. Non-time-dependent threats (manufacturing defects, third-party damage signatures) have less compounding risk because the defect characteristic does not generally worsen between inspections. The asymmetry in the threshold reflects the asymmetry in the consequence.

Coverage is per-segment and per-tool-technology. Different ILI tool technologies detect different mechanisms: MFL tools have well-characterized sensitivity to metal loss, UT tools detect crack-like features, EMAT tools detect axial cracks, caliper tools detect geometry deformation. An operator running combined-technology in-line inspection has a different ITMCI than an operator running MFL only. The metric should be computed per tool technology and then aggregated for the segment, with the aggregation explicit about which mechanisms each technology contributes coverage for.

The edge cases the metric has to handle. The first is small validation sample size for rare mechanisms. The AI vendor may report high recall on selective seam weld corrosion based on a validation set of fifteen verified examples, which is too few to defend a coverage claim. The metric requires a minimum validation sample size per mechanism, proposed minimum is fifty verified examples, below which the claim is "insufficient evidence" rather than "covered" or "not covered." The second edge case is mechanisms applicable to the segment that no ILI tool technology can reliably detect, certain forms of cracking that require hydrostatic pressure testing or in-line crack detection tools beyond what the operator is running. These mechanisms are excluded from the ITMCI denominator and noted separately as inspection gaps that the integrity program needs an alternative assessment for. Carrying them in the denominator would punish the metric for inspection technology limitations rather than for AI scoring limitations, which is not the question ITMCI is asking.

Proposed formula, pending validation pilots.

At a Glance

Category
Pipeline & Midstream Integrity Intelligence
← Corrosion Model Operating Envelope Alignment (CMOEA)Time-Shifted Detection Value (TSDV) →

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