BRENNEN

Industries / Industrial equipment and services / Industrial services / Industrial cleaning

Tank cleaning

Tank cleaning removes the sludge of heavy oil, wax, water and sand that settles on the floors of crude oil and fuel storage tanks, so the tanks can be inspected, repaired or put to other use.

6

Milestones achieved, from 1968 to 2011.

1

To watch for: announced, not yet confirmed outside.

3

Open deliverables. Nobody has shipped them.

In depth

The same ground is charted in more depth, as an industry of its own.

The road

AchievedWatch forOpen

Achieved

  1. 1968

    Tank cleaning gets a guide

    In September 1968 the American Petroleum Institute (API) published the first edition of its Recommended Practice 2015, "Cleaning Petroleum Storage Tanks", a 14-page guide. It set out the hazards of cleaning a petroleum tank and the steps for doing it safely, and it stressed that a safe job starts from knowing the chemical and physical properties of what the tank held.

    firedoc.nist.gov

    Credit

    American Petroleum Institute

  2. 1987

    A cigarette in the tank

    On 11 June 1987 four contractors were cleaning an emptied floating-roof crude oil tank at BP's Dalmeny terminal in Scotland, three of them inside. The roof rested on its 219 support legs, and three roof manholes were opened for natural ventilation; mechanical ventilation and close monitoring of vapor were judged unnecessary. A worker who had taken off his air-line breathing apparatus dropped a lit cigarette, the crude on the floor caught fire, and one man died of asphyxiation and burns. The site rule that matches and lighters be handed in at the gate was not enforced, and the contractors had not been told about the toxic risks or the flammable vapor at the oil surface.

    LessonAn emptied tank still gives off flammable vapor from the oil left on its floor, so ventilation, gas checks and control of flames are needed for the whole job.

    hse.gov.uk

    Credit

    investigated by the UK Health and Safety Executive (HSE), report published 1989

  3. 1987

    Air rules for degassing tanks

    The South Coast Air Quality Management District, the air regulator for the Los Angeles region, adopted Rule 1149 on 4 December 1987 to cut emissions of volatile organic compounds, the evaporating hydrocarbons that help form smog, when storage tanks are cleaned and degassed. Degassing means removing the vapor left inside an emptied tank. As amended in 2008, the rule bars opening a covered tank to the air until the vapor inside stays below 5,000 parts per million, measured as methane, for at least an hour after degassing; requires notice to the district before each degassing; and limits the exhaust of vacuum trucks that remove sludge from such tanks to 500 parts per million.

    aqmd.gov

    Credit

    South Coast Air Quality Management District

  4. 1993

    Cleaning without going inside

    Between 1991 and 1993 Toftejorg of Denmark developed and patented BLABO, a closed system for large crude and fuel oil tanks that dissolves the sludge, pumps it out and separates oil, water and solids outside the tank, so nobody has to enter the tank for the bulk of the work. By 1996 it had been used at more than 50 sites in Scandinavia and northern Europe. Toftejorg's managing director reported 97 to 99% of the hydrocarbons recovered, against 65 to 75% for manual cleaning, and cleaning times of 12 to 16 days for a tank with 1,500 cubic meters of sludge, against 5 to 8 weeks by hand. A 2020 review by Alexandros Chrysalidis and George Kyzas listed BLABO among five cleaning systems on the world market and judged the closed-circuit systems the most economical, able to recover up to 95% of the oil in the sludge. BLABO is now sold by Oreco.

    ogj.comdoi.org

    Credit

    Toftejorg; Jan S. Hummer, its managing director

  5. 1994

    Sonar maps the sludge

    In October 1994 Willacy Oil Services in Britain first filed for a patent, granted in the US in 1999, on lowering a sonar head into the oil of a full tank and sweeping its beam across the floor to measure the shape and volume of the sludge, correcting for the way the speed of sound in the oil changes with depth. Measuring sludge from the roof lets an owner plan a cleaning before the tank is emptied. A 2015 paper compared sludge measurement methods using surveys of more than 30 tanks with external floating, internal floating and fixed roofs, and discussed how accurate each method is.

    patents.google.comdoi.org

    Credit

    George Willacy, Stanislaw Kolaczkowski and Nicholas Pace, Willacy Oil Services

  6. 2011

    Four killed emptying a tank

    On 2 June 2011, at the Pembroke refinery in Wales, then run by Chevron, five contract workers were emptying tank 302 in the amine recovery unit with a vacuum tanker to prepare it for cleaning. A flammable atmosphere had formed inside the tank, and it exploded, most likely ignited by a spark from an unearthed hose. Four workers were killed and a fifth was badly burned; the tank's 5-tonne roof was thrown about 55 m and struck a butane storage sphere. Flammable gas had been checked a few days earlier, but the results were not passed on or not understood. In 2019 Valero, which bought the refinery in August 2011, was fined £5 million plus £1 million in costs, and the cleaning contractor B&A Contracts £120,000 plus £40,000.

    LessonA gas test protects a crew only if its result reaches them and stops the job.

    thechemicalengineer.com

    Credit

    investigated by the UK Health and Safety Executive; reported by The Chemical Engineer from the 2019 court case

Watch for

  1. Robots clean an FPSO's tanks

    SBM Offshore and the environmental services company Ambipar say remotely operated robots removed heavy oily sludge from the cargo oil tanks of an FPSO, a ship-shaped platform that produces and stores oil at sea, while the crew watched around the clock from outside, so no one entered the tanks during the cleaning. They name no vessel and give no figures for time, cost or sludge removed, and a search found no later report of results.

    Achieved whenthe FPSO's operator or an outside party, such as a classification society, publishes a job's sludge removed and days taken and confirms that no one entered the tanks.

    sbmoffshore.com

    Claimed by

    SBM Offshore and Ambipar, August 2025

Open

  1. 01

    Score the sludge forecasts

    A public test, on the same set of tanks, of the ways to estimate how much sludge lies on a crude tank's floor before it is opened: Jiang and co-authors' formula based on wax and asphaltene settling (asphaltenes are the heaviest, tar-like molecules in crude), and a simple model fitted to records the owner already keeps (which crudes the tank held, throughput, temperatures, mixer running hours and years since the last cleaning). Each forecast is scored against a sonar survey or the volume actually removed. It would tell an owner which forecast to trust when choosing between manual and no-entry cleaning.

    The brief
    Credit · unclaimed
  2. 02

    A public tank cleaning log

    A public table of storage tank cleaning jobs drawn from published reports, with the same fields for each: tank size and product, sludge volume, method (manual, no-entry system or robot), days taken, oil recovered, solids sent for disposal, whether anyone entered the tank, and who reported the figures (owner, contractor or equipment maker).

    The brief
    Credit · unclaimed
  3. 03

    Teach a robot to desludge

    A small tracked robot that learns, from a few dozen demonstrations, to push and pump sludge from a tank floor on its own, tested in a mock tank: a shallow steel pan in a workshop filled with a harmless stand-in for sludge, such as sand, clay, water and wax. Nothing about it involves a real tank, oil or a person in a confined space.

    The brief
    Credit · unclaimed

Open 01 · the brief

Score the sludge forecasts

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public test, on the same set of tanks, of the ways to estimate how much sludge lies on a crude tank's floor before it is opened: Jiang and co-authors' formula based on wax and asphaltene settling (asphaltenes are the heaviest, tar-like molecules in crude), and a simple model fitted to records the owner already keeps (which crudes the tank held, throughput, temperatures, mixer running hours and years since the last cleaning). Each forecast is scored against a sonar survey or the volume actually removed. It would tell an owner which forecast to trust when choosing between manual and no-entry cleaning.

Why it is still open

Each published method has been checked at one site. Jiang and co-authors (2021) fitted their formula to four tanks at one pipeline station in China. Petr Straka and Daniel Maxa (2024) built a laboratory model tank at 1:20 scale whose 18-day test matched a year's storage of one Russian crude in one large tank, and found that a crude's composition did not predict how much sludge it formed, so crude assays alone are a weak basis for a forecast. A 2015 comparison across more than 30 tanks found that the accuracy of the surveys themselves is a problem. No comparison of forecasting methods on the same tanks was found.

doi.orgpaliva.vscht.czdoi.org

Built with what exists

The published formulas; Python with pandas and scikit-learn; past sonar surveys or measured removed volumes, shared by tank owners, as the answers to test against.

  1. First step

    Rebuild Jiang and co-authors' formula from their paper, check that it reproduces their four tanks, and ask two terminal operators for ten past cleaning records each.

  2. Done looks like

    At least 20 tanks from at least two owners, each with a survey or a measured removed volume; each method's error published tank by tank, with the share of tanks it gets within 20% of the measured volume; code and anonymized data published.

Open 02 · the brief

A public tank cleaning log

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public table of storage tank cleaning jobs drawn from published reports, with the same fields for each: tank size and product, sludge volume, method (manual, no-entry system or robot), days taken, oil recovered, solids sent for disposal, whether anyone entered the tank, and who reported the figures (owner, contractor or equipment maker).

Why it is still open

Chrysalidis and Kyzas wrote in 2020 that no published work presented the cleaning methods in use, and their comparison of five systems rests on the systems' own descriptions. Job figures that do appear are scattered and uneven: the best-known BLABO results were written up in 1996 by the head of the company that made it; Saudi Aramco's 2021 account of cleaning tank T-814 at Abqaiq gives about 1,800 cubic meters of oil recovered per tank but no sludge volume or days; and a 2024 account of a robot job by the contractor AEKS compares it with an estimated, not measured, manual cleaning time.

doi.orgaramcolife.comtankstorage.com

Built with what exists

Crossref and OpenAlex, free APIs for finding papers; trade magazine archives; a model built for reading long documents, such as Meta's Muse Ember (see the Note "The jobs agents repeat"); Google's Gemini Decision 1 to answer the same fixed questions about each report (see the Note "A thousand questions, one pass"); an agent that cites the page behind each field, such as one built with Cursor's agent kit (see the Note "Findings that cite notes"); a spreadsheet; Zenodo to publish with a DOI.

  1. First step

    Enter the jobs in the three reports cited here, about a dozen, to fix the field list.

  2. Done looks like

    At least 100 jobs from at least 30 sources, each field linked to the page it came from; two people check 20 randomly chosen jobs and agree with the table on at least 95% of fields; a short summary comparing figures reported by equipment makers with those reported by tank owners.

Open 03 · the brief

Teach a robot to desludge

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A small tracked robot that learns, from a few dozen demonstrations, to push and pump sludge from a tank floor on its own, tested in a mock tank: a shallow steel pan in a workshop filled with a harmless stand-in for sludge, such as sand, clay, water and wax. Nothing about it involves a real tank, oil or a person in a confined space.

Why it is still open

Desludging robots in service are driven or watched by an operator: SBM Offshore and Ambipar's robot was monitored by the crew, Re-Gen Robotics calls its system remote controlled, and the robot AEKS uses is remotely operated. Genrobotics lists its sonar-mapping Draco tank robot as a scaled prototype. Work toward autonomy has stopped at following paths: J. G. Hur, Chang Gyoon Lim and Yong Min Kim (2023) gave a robot for sludge in wastewater tanks obstacle avoidance and a path-following function; Zhao, Sun and Lim (2023) planned a desilting robot's path in simulation only; and a hobby project on GitHub (AquaVortex, March 2026) steers a Raspberry Pi sludge robot from Gemini's reading of a camera feed, with no test results. None reports a robot that learned the job from demonstrations and was measured on how much sludge it moved.

tankstorage.comtankstorage.comresearch.genrobotics.comlink.springer.comdoi.orggithub.com

Built with what exists

A hobby-scale tracked chassis with a front scraper or auger and a pump intake; cameras; LeRobot, Hugging Face's open library for recording demonstrations and training robot policies (see the Note "No download"); an open robot model to fine-tune, such as NVIDIA's GR00T 2 (see the Note "Five demonstrations per task") or Runway's Praxis-1 (see the Note "Open robot model weights").

  1. First step

    Build the mock floor and record 20 remote-controlled demonstrations with LeRobot.

  2. Done looks like

    20 filmed runs on a 2 m by 2 m mock floor covered with 10 cm of the stand-in, with nobody touching the controls, in which the robot moves at least 90% of the stand-in by weight to the pump intake; trained on no more than 50 demonstrations; times, failures, code and data published.