Industries / Industrial equipment and services / Industrial services / CMMS and asset management software
Analysis
Analysis turns the maintenance record into decisions: which machines fail, how often and why, and whether the maintenance program is working.
5
Milestones achieved, from 1951 to 2026.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1951
A curve for failures
Waloddi Weibull published "A Statistical Distribution Function of Wide Applicability" in the Journal of Applied Mechanics in 1951. The distribution named after him "is immensely popular in reliability theory, because it includes distributions of decreasing, constant, and increasing failure rates", so one fitted curve shows whether a machine tends to fail young, at random or from wear.
Credit
Waloddi Weibull
1984
Failure data pooled
OREDA was established in 1981 in cooperation with the Norwegian Petroleum Directorate, with the original aim of collecting reliability data on oil industry safety equipment, by Wikipedia's account. Its member energy companies "collect, analyse and exchange" anonymized reliability and maintenance data, by OREDA's own page. The first OREDA Reliability Data Handbook appeared in 1984 and the sixth in 2015. The database is now reached through OREDA@Cloud, paid for by membership fees and by subscriptions from non-members.
Credit
OREDA's member companies; the Norwegian Petroleum Directorate; DNV and SINTEF as project managers
1999
Common failure codes
ISO 14224, first published in 1999 and revised in 2006, sets out how petroleum, petrochemical and natural gas companies collect and exchange reliability and maintenance data. It gives failure modes for each class of equipment and generic lists of failure mechanisms and causes, so data from many sites can be pooled and compared.
Credit
ISO technical committee 67, working group 4
2024
Labelled texts made public
Tyler Bikaun, Tim French, Michael Stewart, Wei Liu and Melinda Hodkiewicz of the University of Western Australia published MaintIE at the LREC-COLING 2024 conference. Two experts independently labelled 1,400 work order texts from mining operations; the 1,067 on which they fully agreed became a public gold-standard set under the MIT license, with a second set of 7,000 texts labelled more coarsely.
LessonTwo experts fully agreed on about three texts in four, which sets the bar any automatic tagger is measured against.
Credit
Bikaun, French, Stewart, Liu and Hodkiewicz
2026
A public test for agents
Dhaval Patel and colleagues at IBM Research built AssetOpsBench, a test of AI agents (programs that use an AI model to plan and carry out a series of steps) on maintenance work. It was first posted in June 2025 and published in the proceedings of the KDD 2026 data-mining conference in August 2026. Its 141 expert-written scenarios run on 4.2K work orders spanning 11 years, taken from IBM Maximo and "enriched with ISO-standard failure codes", and on more than 2.3 million sensor readings from four chillers and two air handling units. The authors report 250+ users and over 500 agents submitted to its public scoring platform.
Credit
Dhaval Patel and co-authors, IBM Research
Watch for
Reports asked for in words
MaintainX says its Report Builder AI "transforms asset and work data into custom reports and visualizations using natural language", drawing on work orders, parts use, purchase orders, schedules and asset history. The release gives no accuracy figures.
Achieved whenan outside test checks the tool's answers against the same reports built by hand.
Claimed by
MaintainX, March 2026
Open
01
Check the report answers
A public test of AI tools that answer questions about maintenance data: a realistic CMMS database holding two years of work orders, parts and meter readings, 200 questions whose answers are computed exactly, and a tally of each tool's right, wrong and refused answers.
The briefCredit · unclaimed02
An open KPI calculator
Open-source code that computes the standard maintenance key performance indicators (KPIs), such as the share of planned work, schedule compliance and mean time between failures, from the exports of common CMMS products, each formula tied to the SMRP metric or EN 15341 indicator it follows.
The briefCredit · unclaimed
Open 01 · the brief
Check the report answers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of AI tools that answer questions about maintenance data: a realistic CMMS database holding two years of work orders, parts and meter readings, 200 questions whose answers are computed exactly, and a tally of each tool's right, wrong and refused answers.
Why it is still open
Plain-language reporting now ships in CMMS products without accuracy figures. AssetOpsBench scores research agents on IBM's own data and does not test the features that ship in commercial products, and searches found no public test of those against a known database.
Built with what exists
Atlas CMMS (AGPLv3) loaded with generated data; SQL queries for the answer key; AssetOpsBench's scenarios and scoring code (Apache 2.0 license) as a model for the questions; the reporting features of shipping CMMS products; an AI agent given read access to the database, built for example with the Cursor Agent SDK (see the Note "Findings that cite notes").
First step
Generate one year of work orders for 50 assets in Atlas CMMS and write 30 questions with their SQL answers.
Done looks like
200 questions, from mean time between failures by asset to parts used by month, scored for at least three tools, with every wrong answer and its cause published.
Open 02 · the brief
An open KPI calculator
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Open-source code that computes the standard maintenance key performance indicators (KPIs), such as the share of planned work, schedule compliance and mean time between failures, from the exports of common CMMS products, each formula tied to the SMRP metric or EN 15341 indicator it follows.
Why it is still open
The Society for Maintenance & Reliability Professionals (SMRP) publishes "70+ metrics and guidelines" in a document "available free of charge to members", so each organization codes the formulas itself. Searches found no open code that computes them from CMMS data.
Built with what exists
Python and pandas; exports from Atlas CMMS, openMAINT and the CSV exports of commercial products; unit tests built from worked examples.
First step
Code five indicators against one CMMS export and check them by hand.
Done looks like
At least 20 indicators that match hand calculations on three test datasets, then used to publish the indicators for one real organization's year of data.