Industries / Energy, water and waste / Pipelines and midstream
Control rooms
Control rooms watch and run a pipeline system from one place, using SCADA (supervisory control and data acquisition), software that gathers readings and sends commands to distant pumps and valves.
5
Milestones achieved, from 1968 to 2021.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1968
Computers run a pipeline
In 1968 IPL Energy became "one of the first pipeline companies in the world to implement computer based control" of its lines, according to a 1998 paper on its SCADA system, which the company kept improving.
Credit
IPL Energy
1999
The control system stalls
On 10 June 1999 a 16-inch Olympic Pipe Line gasoline line ruptured in Bellingham, Washington, and about 237,000 gallons ran into a creek and caught fire, killing two 10-year-old boys and an 18-year-old. The NTSB named five causes, including earlier damage by a contractor, and the company's practice of developing the SCADA database on the live system, which left it unresponsive at a critical time.
Lessonnever change the control system while it runs the line.
Credit
Olympic Pipe Line Company (operator); NTSB (investigation)
2009
Rules for control rooms
A US rule of 3 December 2009 requires operators whose controllers use SCADA to define controller authority, follow API practice RP 1165 for SCADA displays, limit hours to manage fatigue, keep a written alarm plan with monthly review of problem alarms, and train controllers on abnormal events, including by simulator or table-top exercise.
Credit
PHMSA
2010
Seventeen hours unseen
Enbridge's Line 6B ruptured at Marshall, Michigan, during a planned shutdown. Leak and SCADA alarms fitted a rupture, but staff took them for column separation (a partly empty line), restarted twice, and pumped in 81% of the total spill before an outside caller reported oil more than 17 hours later.
Lessonwhen alarms and a flow imbalance agree, assume a leak first.
Credit
Enbridge (operator); NTSB (investigation)
2021
Ransomware stops a pipeline
Just before 5 a.m. on 7 May 2021 a Colonial Pipeline employee found a ransom note on the company's business network. To keep the malware off the operational network that runs the pipeline, the control center shut all 5,500 miles in about 15 minutes. Lines began returning to service on the evening of 12 May. The attacker used an old VPN (remote access) profile that was not meant to be in use.
Lessonif the business and control networks cannot be shown to be separate, a breach of one can stop the other.
Credit
Colonial Pipeline; testimony by CEO Joseph Blount to the US Senate, 8 June 2021
Watch for
An AI aid for operators
ABB says its Augmented Operator software spots small changes in a process, searches years of plant history for similar cases, forecasts key readings and simulates the effect of a change, while leaving decisions to the operator. ABB says it can reduce unplanned shutdowns and gives no figures.
Achieved whenan operator publishes measured results, such as fewer missed abnormal events in a test or in service, checked by someone outside ABB.
Claimed by
ABB, November 2026
Open
01
Replay Marshall for an assistant
A test of whether an AI assistant, reading the same alarms and readings a controller sees, can tell a rupture from column separation during shutdowns and restarts.
The briefCredit · unclaimed02
Turn accident reports into timelines
Machine-readable timelines (time, reading or alarm, controller action) for the main control-room accidents, taken from public NTSB reports.
The briefCredit · unclaimed
Open 01 · the brief
Replay Marshall for an assistant
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of whether an AI assistant, reading the same alarms and readings a controller sees, can tell a rupture from column separation during shutdowns and restarts.
Why it is still open
At Marshall, staff took rupture alarms for column separation and pumped oil into the broken line twice. A search for published tests of language models on this problem found none.
Built with what exists
The simulated runs from "A software leak test bench" above; language models, including small ones that run on one machine (Notes: "Checked without being asked", "A first stable number for local AI"); Python.
First step
Write ten shutdown and restart scenarios as streams of alarms and readings, five with a rupture.
Done looks like
Across at least 200 scripted runs, half with a rupture, the share of ruptures flagged before a restart and the false-alarm rate, published with prompts and code; a second group reruns them.
Open 02 · the brief
Turn accident reports into timelines
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Machine-readable timelines (time, reading or alarm, controller action) for the main control-room accidents, taken from public NTSB reports.
Why it is still open
The NTSB reports tell these sequences in prose, for example the Marshall alarms and restarts. No machine-readable set was found, and such timelines are the raw material for controller training and for tests like the one above.
Built with what exists
NTSB reports PAR-12/01, PAR-02/02 and PAR-11/01; an AI model that cites the page behind each row (Note: "Findings that cite notes"); a spreadsheet.
First step
Build the Marshall timeline for 25 and 26 July 2010 from the report's narrative.
Done looks like
Timelines for at least three accidents, every row with a page reference; a second person checks every row of one timeline.