Introduction

Inspection technology has advanced considerably over the last several years, and each year these advancements are presented at API’s Inspection & Mechanical Integrity Summit. While the developments are promising, their rate of success lies heavily with the data these tools rely on. This article recounts several presentations from recent API Summits and examines common data challenges, their impact on inspection technology, and the factors to consider before the next related software purchase or pilot.

Where Inspection Technology is Working

The most visible advancement in inspection technology is in data collection. Drones, crawlers, and robotic tank tools now place sensors where scaffolding used to go. Phased array, total focusing method, guided wave, and pulsed eddy current detect damage that a grid of spot readings can miss, and permanently installed and wireless thickness sensors replace individual site visits with a continuous record.

Data management has changed more quietly. Inspection Data Management Systems (IDMS) and corrosion management platforms and digital isometrics and P&IDs convert drawings and spreadsheets to structured records that can be readily referenced and used.

On the decision side, risk-based inspection (RBI) software, damage mechanism screening, analytics, and AI-assisted review prioritize and flag inspections across thousands of circuits in a fraction of the time traditionally required. Execution tools, including mobile field apps, digital work packages, and turnaround planning software, deliver the inspection plan to the person holding the gauge and then return the result.

The technology used for data collection, data management, decision-making, and execution all consume or produce the same set of records: circuits, corrosion monitoring locations (CMLs), damage mechanism assignments, thickness history, and open recommendations. This shared dependence is where the trouble starts.

Where Inspection Technology Stalls

While current inspection technology is intended to advance inspection programs, four common data failures can undermine the results delivered by these tools:

1. The circuit is drawn wrong. RBI and corrosion control documents operate at the circuit level, so the circuit boundary determines which assets are analyzed together. A session from ISQ Brasil at the 2026 API Inspection & Mechanical Integrity Summit presented four cases in which changing how the equipment was divided changed the result. Splitting a gas separator by active damage mechanism and inspection accessibility put its segments in different risk cells, 5C and 3C. Refining the segmentation of a two-stage exchanger produced risk cells 5D and 4D. A circuit drawn too coarsely hides its worst segment within the average, and the inspection plan then targets the average.
2. The measurement is wrong and nobody knows. A fired heater session compared two infrared camera and technician combinations on a single tube. The raw readings were 894°F (478.9°C) and 796°F (424.4°C). After correction for emissivity and reflected background, they were 716°F (380°C) and 706°F (374.4°C), within a calculated uncertainty of about ±44°F (6.7°C) and ±39°F (3.9°C), whereas conventional correction left them at 860°F (460°C) and 745°F (396.1°C). Tube metal temperature feeds the integrity operating window and the remaining life estimate, and an error of that size at the source can cascade downstream. A better instrument does not help when the resulting data has a wide range of uncertainty.
3. The inspection is credited with more than it can see. In a session on RBI blind spots, an HF alkylation carbon steel circuit was rated low probability of failure based on a corrosion rate under 0.03 mm (0.001 in)/year from UT history. Inspection effectiveness was scored high because UT was performed on the RBI schedule. The line ruptured at the 6 o’clock position midway along an elevated 5 m (0.197 in) horizontal run. The damage was highly localized, and the adjacent areas showed no measurable wall loss. The monitoring points showed no corrosion because the damage was not at the monitoring points, and the high effectiveness score precluded consideration of a screening method such as guided wave. A second case in the same session turned on an assumption: a sulfur degassing unit was monitored for water dewpoint and treated as dry although acid condensation forms at much higher temperatures. The carbon steel piping failed within weeks, and the C-276 replacement failed after 12 months. The presenters concluded that RBI is only as good as the data and execution behind it.
4. The finding never reaches the plan. A Summit session on mechanical integrity at small sites cited incidents wherein the common thread was inspection recommendations that were never completed, including a 2023 reboiler failure that led to a 47-day shutdown.

Do you know what data a decision needs? Can you reach it across silos, PDFs, and inconsistent naming? Can you tell which of it is current and correct?

Case Study: One CML, Three Times the Rate – Why Clean Data Made the Finding Believable

The Setting

In a pipe still unit at a North American refinery, a circuit in caustic and butane service was found to have localized corrosion. It was one of 167 circuits, with 6,900 CMLs and 27,000 thickness measurements in this unit’s piping integrity program. We used CorrSolutions’ patented Bayesian Optimized Asset Reliability (BOAR) analysis to identify areas susceptible to localized corrosion. Damage mechanism wording was then added to the isometrics so inspectors could see where localized corrosion was suspected, as well as where localized corrosion had already been found.

Before

Circuits were inspected based on traditional CML quantity calculations, which included additional CMLs when localized internal corrosion was the applicable damage mode. Since a thickness projection is only as good as its starting point, we included a review of suspect nominal thicknesses provided by the client IDMS.

The review turned potentially misleading projections into actionable recommendations, and it exposed one CML with a genuine corrosion rate of 9 mpy. Relying on the circuit’s average corrosion rate of 0.7 mpy could have meant this location was lost in the volume of data being managed. In another example, a circuit being analyzed with as-received nominal thicknesses yielded a falsely high corrosion rate of 4.8-12.9 mpy. After data cleanup, the corrosion rate was recalculated as 1.6-6.6 mpy. A projection built on the wrong data would have misjudged these locations in either direction, burying a problem or masking the location as one more bad reading in a noisy dataset.

The Lesson

Proper analytics rewards clean inputs and can turn noisy data into accurate recommendations. Best-in-class analytics can assist in both the data cleanup, and analysis used in inspection planning.

Considerations with Inspection Technology

Six strategic considerations can help avoid the pitfalls described above:

1. Can you trust the data under the tool? Before the next purchase, audit what the tool will be reading. Pull a sample of circuits and check the boundaries against the damage mechanisms, compare CML locations with the isometrics, and count the open recommendations while noting the age of the oldest. A few days spent on these three checks will indicate more about whether a new tool will pay off than the vendor demonstration will.
2. Which decision does the tool change? Identify the interval, scope, or replacement decision that the tool will inform. A tool that cannot be tied to one of these decisions adds only a dashboard.
3. Who owns the output? Separate facts from interpretation, link each result to its source, and record the reviewer.
4. Where are results stored? Results belong in the system of record. One 3D digital twin session at the summit made this point directly: the twin supplies spatial context, and the IDMS remains the system of record.
5. What happens when results are wrong? Decide in advance how an erroneous reading, an incorrect circuit definition, or a bad model output will be caught, as well as who will be notified. In the spot-UT case, the absence of QA/QC meant the first indication of the error was a rupture.
6. Start narrow. Equity Engineering’s Summit session with Koch Ag & Energy Solutions recommended beginning with high-risk assets and proven use cases, then expanding only when results are well-documented and repeatable.

Where Inspection Technology is Heading

Routine review work is the most likely to be automated first. Vendors at API’s 2026 Summit demonstrated artificial intelligence (AI) platforms that validate NDE data, detect and size indications, and draft the report. One speaker argued that AI will gain wide acceptance once agentic workflows reduce operating and capital cost, partly by allowing less senior staff to perform work that now requires high-end consultants. While this may reduce cost, it does not resolve the question raised by the closing panel of the 2024 Summit: who watches the watcher?

Our position is straightforward: large language models (LLMs) and analytics should retrieve, cross-check, and draft, while engineers review and decide.

Controls on accuracy, version, and security must be in place before the tool is trusted, because a model that reads outdated records will repeat them with confidence. Every tool in this article amplifies the assumptions it is given, whether good or bad. As routine reviews move to software, the engineer’s time shifts toward the judgments that software should not make: whether a circuit boundary is correct, whether an effectiveness rating was earned, and what a surprising reading means.

References

• Data Graveyard to Decision Engine (Brandt), 2026 Inspection Summit.
• The Cold Truth About Hot Data (Anam Rizvi, Quest Integrity), 2026 Inspection Summit.
• RBI Blind Spots: Lessons from Real-World Failures (Mei Ling Cheah, Veronica Dominguez, Jozef Soltis), 2026 Inspection Summit.
• Managing integrity in corrosion circuits (Fernanda Monteiro, Rodolpho Delunardo, Vitor Limongi, ISQ Brasil), 2026 Inspection Summit.
• From Buzzwords to Breakthroughs (Juliet Calleton, EEG; Albert Suckow, Koch Ag & Energy Solutions), 2026 Inspection Summit.
• From Isos to Insights (Ryan Streeter, MISTRAS; Eric Hoffman, IMS), 2026 Inspection Summit.
• MI on a Budget (Lisa Czyszczewski and Chase Hauser, MISTRAS), 2026 Inspection Summit.
• AI-Driven Transformation of the Inspection Workflow (Oleg Volf, AUToSCAN-AI), 2026 Inspection Summit.
• AI acceptance, not until automation of work processes (Jarret Reeves, AsInt), 2026 Inspection Summit.
• API Summit Closing Panel, “Ask the Experts” and the Future for Mechanical Integrity, January 2024.