Industries / Industrial equipment and services / Industrial machinery / Boilers and pressure vessels
Inspection and repair in service
This work checks boilers and vessels at intervals once they are running, judges whether damage found can be left in service, and repairs them under rules that keep the repair as sound as the original.
7
Milestones achieved, from 1945 to 2017.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1945
Rules for repairs
The National Board published the first chapter of the National Board Inspection Code (NBIC) in 1945, giving rules for inspecting, repairing and altering boilers and vessels after they leave the shop; it became an American national standard in 1983. In 1975 the Board introduced the "R" certificate, which authorizes a firm to repair code vessels.
Credit
National Board of Boiler and Pressure Vessel Inspectors
1958
An inspection code for refineries
The American Petroleum Institute (API) first published API 510, its code for inspecting pressure vessels in service, in 1958, carrying on rules from a joint API/ASME code of 1934 that ended in 1956. Its 2006 edition lets owners set inspection intervals by risk-based inspection under API 580 and 581, which ranks each vessel by how likely it is to fail and how bad a failure would be, and accepts fitness-for-service assessment under API 579.
Credit
American Petroleum Institute
1977
Repairs without heat treatment
Heat treatment after welding is hard to do on a vessel in place. TWI says the 1977 NBIC was the first to include weld repair methods for which post-weld heat treatment could be left out, for carbon and some low-alloy steels, using controlled ways of laying the weld beads such as the temper bead method, in which each layer tempers the one before. Cr-Mo steels were added in 1995 and some others in 2007.
Credit
National Board; technical basis in W. D. Doty's Welding Research Council Bulletin 412 (1996)
2000
A written scheme for each system
The UK Pressure Systems Safety Regulations came into force on 21 February 2000. A user may not operate a pressure system without a written scheme for its periodic examination by a competent person, and HSE says an in-house competent person must have the authority to stop its use. They cover steam at any pressure and gases above 0.5 bar above atmospheric.
Credit
UK Parliament; Health and Safety Executive
2007
Fit for service
A fitness-for-service assessment decides by calculation whether a vessel with corrosion, cracks or dents can keep running, and for how long. API published API 579 in 2000, and a joint API and ASME committee replaced it in 2007 with API 579-1/ASME FFS-1, adding parts on creep, dents and laminations. A manual of worked examples, API 579-2/ASME FFS-2, followed in 2009.
Credit
American Petroleum Institute; ASME
2010
Damage inspection cannot find
In April 2010 a forty-year-old heat exchanger ruptured at the Tesoro refinery in Anacortes, Washington, causing a fatal explosion and fire. The US Chemical Safety Board (CSB) found the cause was high temperature hydrogen attack, in which hydrogen at high temperature cracks carbon steel from within, and concluded that "inspections should not be relied on" to find it because the cracks are localized and few inspectors have the skill. CSB asked API to ban carbon steel in such conditions, and rated API's 2016 response "Closed – Unacceptable Action".
LessonWhere inspection cannot reliably find damage, the safeguard has to be the choice of material.
Credit
US Chemical Safety Board
2017
A repair that left thin steel
On 3 April 2017 a steam system vessel at Loy-Lange Box Company in St. Louis failed at its bottom head and was thrown through a neighbouring roof, killing one employee and three members of the public. CSB found oxygen pitting had thinned its 0.250 in wall to as little as 0.012 in, and that a 2012 repair by an R-certificate holder likely left steel thinner than the required 0.2204 in, which the inspector accepted. CSB asked the National Board to add rules that would stop inspectors accepting such repairs.
LessonA repair must remove all the damaged metal, and both repairer and inspector have to measure that it has.
Credit
US Chemical Safety Board, report of August 2022
Watch for
AI estimate of boiler damage
Visionaize says V-BTHERM, an AI model, tracks creep and fatigue damage in boiler parts from live plant control data and can support remaining-life assessments. It names no customers and gives no accuracy figures.
Achieved whena named utility publishes the model's predicted damage against damage measured at an outage, for example by metallographic replicas.
Claimed by
Visionaize, November 2026
A faster inspection crawler
Gecko Robotics says its Raptor robot, which climbs steel to measure wall thickness by ultrasound, collects data three to five times faster than its earlier robots, with ADNOC as first customer; it names pipelines, storage tanks and steel structures, not boilers.
Achieved whenADNOC or an independent party publishes field results comparing Raptor's coverage and thickness readings with manual ultrasonic testing.
Claimed by
Gecko Robotics, November 2026
Open
01
Open boiler tube thickness maps
A public, licensed data set of ultrasonic thickness maps of boiler furnace wall tubes, with tube position, elevation, nominal thickness and a label for thinned areas, and a baseline that finds thinning and is scored against those labels.
The briefCredit · unclaimed02
Fitness-for-service tools, checked
A public test suite that runs the worked examples for general and local metal loss in the 2009 API 579-2/ASME FFS-2 manual against the open-source fitness-for-service calculators, and reports where each matches and where it does not.
The briefCredit · unclaimed
Open 01 · the brief
Open boiler tube thickness maps
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, licensed data set of ultrasonic thickness maps of boiler furnace wall tubes, with tube position, elevation, nominal thickness and a label for thinned areas, and a baseline that finds thinning and is scored against those labels.
Why it is still open
The one public example found, agurani's "CFB-Boiler-Tube-Mapping" on GitHub, describes measurements across 480 tubes per level but publishes one spreadsheet and some images, with no license. A search of GitHub, Zenodo and Kaggle in November 2026 found no other data set of boiler tube thickness.
Built with what exists
Thickness records given by plant owners with permission, or tubes cut out at outages and measured in a laboratory; Python with pandas; Zenodo for publication.
First step
Ask the owner of the GitHub repository for a license and the full set of files, and ask one university with a power plant for its outage records.
Done looks like
At least 100,000 thickness points from at least two boilers under an open license, and a baseline thinning detector with its precision and recall published.
Open 02 · the brief
Fitness-for-service tools, checked
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test suite that runs the worked examples for general and local metal loss in the 2009 API 579-2/ASME FFS-2 manual against the open-source fitness-for-service calculators, and reports where each matches and where it does not.
Why it is still open
ASME publishes the 2009 example manual free on its site, with worked problems for general and local metal loss. Open tools exist, among them flare9x's FitnessForService.jl (MIT license, last updated March 2021), which tests its own local metal loss routines, and AbundantSalmon's api-579-calculator, which covers one Level 1 case. No shared suite runs the same examples across them.
Built with what exists
The free manual's input and output values, cited rather than copied; Julia and Python test frameworks; GitHub Actions.
First step
Turn the first general metal loss example into a test case for FitnessForService.jl and open a pull request.
Done looks like
All the manual's general and local metal loss examples reproduced within a stated tolerance in at least two packages, run automatically, with each difference reported to the package's authors.