Day 2 of IrthBound 2026, our user summit earlier this month, opened with a panel on how integrity programs are evolving. The theme was easy to state and much harder to live: regulations and failures are forcing integrity programs to evolve. Pipelines are aging, threats are more complex, and traditional assessment methods aren't enough. Operators are conducting more inspections and running more technologies, which ultimately means more data. The question our panelists kept coming back to was how you turn that data into decisions.
One of our panelists described how his team is moving from deterministic to probabilistic analysis for corrosion, crack, and risk to run a more efficient integrity program. Probabilistic analysis came up again later on Day 2, when Irth's asset integrity team presented the probabilistic corrosion growth rate (CGR) model we've been developing for Asset Integrity for Pipelines (AIP). That session generated some of the most interest of any asset integrity session at the summit.
That level of interest tells us something about where the industry is right now. Most integrity teams we talk to still rely on some form of deterministic growth (i.e., a single rate per anomaly or per segment), and most of them already know its limitations. The harder questions are what to replace it with, how to defend that choice, and whether the alternatives are practical across a full system's worth of inline inspection (ILI) data. Below is a recap of what the session covered for those who couldn't be in the room, and for those who were and want to dig back in.
The most common approach to corrosion growth is still anomaly-to-anomaly, if you will: match a metal loss feature between two ILI runs, subtract the earlier reported depth from the later one, and divide by the years between inspections. It's simple and easy to explain, but the resulting rate is heavily influenced by tool error and tolerance, especially when the true growth between runs is small relative to the sizing uncertainty on each call. In practice, this can result in assigning the fastest CGR to the shallowest anomaly and the slowest CGR to the deepest anomaly, which is exactly backward from where an integrity engineer wants conservatism. It also means the effective safety factor varies from one anomaly to the next, so two features with the same reported depth can end up on very different repair timelines for reasons that have more to do with the tool than with the pipe.
If the tool error between two runs is as large as the growth you're trying to measure, what's that two-point growth rate actually telling you?
None of this is a new concern. In July 2009, PRCI published Development of Detailed Procedures for Comparing Successive ILI Runs to Establish Corrosion Growth Rates (Contract PR-331-063525), a user's guide prepared by GE and C-FER Technologies. At the time, there was no industry guidance on how to compare repeat ILI data sets, even as more and more pipelines were being inspected for a second, third, or fourth time. The report laid out a six-step process (i.e., obtain the ILI data, select a comparison method, check weld alignment and defect consistency, evaluate data uncertainties, determine growth rates, and check them against expected values), along with Microsoft Excel templates to work through it.
Its central point still holds up: because the actual size of a defect is uncertain given its measured size, the growth rates calculated from those sizes are also uncertain and are best characterized probabilistically. The report also noted that the ratio between measured corrosion growth and measurement error determines whether a growth rate distribution means anything at all; when that ratio is small, the uncertainty can be too large to make meaningful probabilistic inferences. It compared local growth estimation (i.e., matching individual defects or signals between runs) with segment growth estimation (i.e., comparing the defect population within a segment). It suggested that a combination of the two could provide the best result, identifying fast-growing features while averaging out measurement error across a segment.
The Irth session framed the evolution of corrosion growth modeling since then in four stages, each one answering a slightly different question:
The progression moves from estimating a growth rate, to accounting for ILI uncertainty, to modeling growth probabilistically, and finally to predicting future depth analytically. Each step trades a little simplicity for a much more honest picture of what the data can and can't tell you.
Monte Carlo simulation is powerful, but it's also computationally heavy. As we discussed when we presented this work at PRCI REX 2025, more complex probabilistic models can prove inefficient or even impossible to run depending on the number of matched anomalies, and a full system can easily involve hundreds of thousands of them.
Irth's model takes a different route. It's an analytical, N-point solution that uses the full ILI depth history for each matched anomaly, along with the measurement uncertainty of each run, to generate a probability distribution of both future depth and growth rate directly, without simulation. Because every run is used, each additional ILI run reduces uncertainty instead of being set aside in favor of the two most recent inspections.
If you've paid for five ILI runs on a pipeline, why would your growth model only use two of them?
A probability distribution of future depth is also what an operator needs to calculate probability of exceedance (PoE), i.e., the likelihood that an anomaly will cross a depth or pressure limit state in a given year. That matters for defensibility, since PHMSA's Liquid IM FAQ 7.19 states that it is acceptable to use probability of exceedance to establish response criteria.
What makes any multi-run growth model practical? Alignment, specifically weld alignment and feature matching across every ILI run on a line, regardless of vendor or tool type. A model that uses the full ILI history is only as good as the matched history underneath it. AIP ingests ILI data sets, performs weld alignment and feature matching per spool, and maintains feature lineage across runs, so the matched history is already in place when the growth analysis starts. Our panelists made the same point earlier in the day: aligning all of their assessment history so they can compare runs is where a normalized database starts turning data into information.
The Irth session also shared a look at the proof of concept underway in AIP, which pairs the probabilistic CGR output with probability of failure, along with a customer use case. What if the probability of exceedance for every matched anomaly were recalculated automatically as soon as a new ILI run is aligned in AIP? That's the direction we're heading, and the feedback from the summit is shaping what comes next.
Working with operators across AIP, we've noticed that operators calculate corrosion growth in many different ways. The variation shows up in the calculation itself (e.g., half-life or percent life from a single ILI, two-point or pit-to-pit between two ILIs, a trend through three or more runs, a joint or segment rate at the maximum or 95th percentile, a probabilistic model, or a fixed default rate). It also shows up in the methodology choices around the edge cases, which can move the final number just as much. When a matched anomaly shows zero or negative growth, some operators apply zero, some apply a fixed default (e.g., 12 or 16 mpy), some apply a minimum floor, and others go to half-life or engineering review. When an anomaly is unmatched, some assume it's new since the last ILI and grew from zero, some assume it was at the reporting threshold at the last ILI, and some apply half-life, full life, or the joint or segment rate. AIP now supports 108 combinations of base methods and edge-case options.
If two operators run the same ILI data through two different methods and get two different dig lists, which one is right?
Our latest release, AIP 3.53, includes several changes aimed at giving integrity engineers more control over exactly these choices:
Flexibility in the software only goes so far if the industry can't agree on which method to use in which situation. That's why Irth is proposing a new research idea to PRCI's Integrity & Inspection Technical Committee at the upcoming Fall Committee Meeting in Calgary: Benchmarking Corrosion Growth Rate Methods to Determine Accuracy and Scenario Dependency (Idea 5113). The idea builds directly on PR-331 and on the N-point probabilistic extension we presented at PRCI REX 2025, and it complements PRCI's NDE-4-30 work on a common ILI data standard format, which improves the run-to-run matching inputs this project would evaluate.
The proposal would test the main growth methods, from simple single rates to probabilistic models, against one anonymized dataset from multiple operators and score each of them the same way, including how accurately it predicts depth and whether it picks the right features to dig. The goal is to answer a simple question with real data: does one method perform best, or does the best method depend on the situation (e.g., number of ILI runs, a change in vendor, or pipe vintage)? The result would be practical guidance that helps engineers choose a growth method based on how it actually performs on data like theirs, rather than by default or habit. Rising tides lift all ships, and this is exactly the kind of question the industry is better off answering together.
If your team is evaluating a move away from deterministic growth, or you want to see how probabilistic CGR could work with your own ILI history in AIP, we'd love to talk. Reach out to sales@irthsolutions.com to set up a conversation or to request a copy of the IrthBound 2026 presentations. If your company is a PRCI member and interested in supporting the benchmarking proposal, we'd love to hear from you as well. We hope to see you at next year's Irth User Summit.