BRENNEN

Industries / Industrial equipment and services / Industrial machinery / Compressors

Controls

Matching a compressor's output to what the plant is using, and keeping the machine out of conditions that damage it.

4

Milestones achieved, from 1974 to 2016.

3

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1974

    Controllers built to stop surge

    Surge is a flow reversal in a centrifugal compressor that comes at low flow and can damage the machine. It is avoided by keeping flow above a limit, if need be by blowing air off or returning it to the inlet, as DOE's sourcebook explains. Compressor Controls Corporation (CCC), based in Des Moines, Iowa, was founded in 1974 on the patents of its co-founder Naum Staroselsky to make controllers for this job; CCC says the control methods before it were pneumatic. When Honeywell agreed to buy CCC for $670 million in April 2023, it said CCC had an installed base of more than 14,000 control applications and served mainly LNG, gas processing, refining and petrochemical plants.

    process.honeywell.comhoneywell.comenergy.gov

    Credit

    Naum Staroselsky and co-founders, Compressor Controls Corporation

  2. 1994

    Motor speed follows demand

    A fixed-speed screw compressor meets falling demand by unloading, and while unloaded it still draws 15 to 35% of its full-load power, by DOE's figures. A variable speed drive (VSD) slows the motor instead, so power falls almost in step with output. Atlas Copco launched its VSD compressors in 1994 after a prototype finished in the late 1980s; it says it was not the first company to offer the technology but brought it to market widely. DOE's 2020 federal standard gives variable-speed rotary compressors their own equipment classes.

    LessonA VSD saves only at part load; near full output it uses more power than a fixed-speed machine, because the drive itself loses energy, the sourcebook notes.

    atlascopcogroup.comenergy.govfederalregister.gov

    Credit

    Atlas Copco

  3. 1998

    A computer runs the room

    Kaeser built Sigma Control, a controller based on an industrial PC, into its rotary screw compressors from 1998. In 2001 it launched Sigma Air Manager, a master controller that runs all the compressors in a station over a network, by Kaeser's own history. DOE's sourcebook describes this class of "system master controls": they can run compressors of every brand and type, with dryers and storage, and log trends for maintenance.

    kaeser.comenergy.gov

    Credit

    Kaeser Kompressoren

  4. 2016

    Networks tied to one brand

    The third edition of DOE's sourcebook, published in March 2016 with the Compressed Air Challenge, warns that network controls built by linking each compressor's own controller usually work only with compressors of the same brand and configuration, because the controllers cannot talk to others, and that upgrades can be costly or unavailable. It also says a variable-speed compressor in a group should always be the trim machine, the one left running part-loaded while the others run full.

    LessonControls that cannot talk across brands tie a plant to one maker or to a separate master controller.

    energy.govosti.gov

    Credit

    US Department of Energy Advanced Manufacturing Office; Compressed Air Challenge; Lawrence Berkeley National Laboratory; Resource Dynamics Corporation

Watch for

  1. Demand matching for fixed speed

    ELGi says its Demand=Match system lets a fixed-speed compressor follow plant demand by recirculating air inside the compressor, giving "up to 17% energy reduction in typical operating profiles". The published release still carries an editing note after that figure: "field test results — confirm with Product/Testing".

    Achieved whenan outside party publishes measured energy per unit of air before and after fitting it at named plants.

    itvoice.inthemanufacturingfrontier.com

    Claimed by

    ELGi Equipments, September 2025

  2. AI set points for rooms

    CompAir says Ecoplant, a compressed-air management system run by machine learning, adjusts compressor set points and chooses which machines run, cutting energy costs and CO2 emissions by up to 30%. No site or baseline is named.

    Achieved whena utility or independent auditor reports measured savings at named sites against a stated baseline.

    ognnews.com

    Claimed by

    CompAir, August 2025

  3. Learned control claims 4% savings

    Vincent Bezold and four co-authors report that a reinforcement-learning policy (software that learns its settings by trial) kept a compressed air system within its limits, cut unneeded overpressure and saved about 4% of energy compared with the installed industrial controller, across several compressor configurations.

    Achieved whenthe work passes peer review and a plant runs the policy in closed loop with measured savings reported by someone other than the authors.

    arxiv.org

    Claimed by

    Bezold, Wagner, Hofmann, Huber and Sauer (arXiv preprint), December 2025

Open

  1. 01

    A shared test for controllers

    A public benchmark for compressor-room controllers. It holds at least 20 week-long logs from real plants (each compressor's load state and power, discharge and header pressure, and flow where metered, every 1 to 10 seconds), the compressors' rated data, an open simulator that replays each plant's demand against a chosen control strategy, and a script that scores energy per unit of air and minutes below the plant's minimum pressure.

    The brief
    Credit · unclaimed
  2. 02

    Surge recordings anyone can use

    An open, labeled set of recordings of small centrifugal blowers throttled slowly into surge, with outlet pressure, sound and vibration logged together, and a baseline detector that warns before surge begins.

    The brief
    Credit · unclaimed

Open 01 · the brief

A shared test for controllers

Credit · unclaimed

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

What to build

A public benchmark for compressor-room controllers. It holds at least 20 week-long logs from real plants (each compressor's load state and power, discharge and header pressure, and flow where metered, every 1 to 10 seconds), the compressors' rated data, an open simulator that replays each plant's demand against a chosen control strategy, and a script that scores energy per unit of air and minutes below the plant's minimum pressure.

Why it is still open

New controllers are each scored only against their own plant's existing controller. A reinforcement-learning study reports about 4% savings and a 2026 predictive-control study estimates 2.8%; neither abstract mentions released data, so the two cannot be compared. The only open compressed-air operating data found on Zenodo is one Chinese industrial park's network flow and pressure, with no compressor-level records, and the best-known open compressor set, MetroPT-3, logs one compressor on a metro train in Porto for failure prediction, not control. DOE's open MEASUR software models compressed air systems but ships no shared demand logs.

arxiv.orgarxiv.orgzenodo.orgarchive.ics.uci.edugithub.com

Built with what exists

Data exports from the controllers plants already have, or clamp-on power loggers and pressure transducers; makers' CAGI data sheets for compressor ratings; the compressed-air module of MEASUR (Oak Ridge National Laboratory, open source); Python with pandas for the replay and scoring.

  1. First step

    Ask three plants that already log compressor data, for example through a utility efficiency program, for one week each, and write the replay script against the first.

  2. Done looks like

    20 plant-weeks released under an open license; when fed each plant's own controller actions, the simulator reproduces that week's measured compressor energy within 5%; three baseline strategies (cascaded set points, a single set point with a VSD trim machine, and one published method) scored on all 20.

Open 02 · the brief

Surge recordings anyone can use

Credit · unclaimed

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

What to build

An open, labeled set of recordings of small centrifugal blowers throttled slowly into surge, with outlet pressure, sound and vibration logged together, and a baseline detector that warns before surge begins.

Why it is still open

Experimental surge work keeps its signals in the lab. The University of Genoa recorded the acoustic and vibration response of two automotive turbocharger compressors in incipient surge, and no download was found. The one open surge classifier found on GitHub trains on 4,000 synthetic points, and a Zenodo search for compressor surge returned modeling papers only.

saemobilus.sae.orggithub.comzenodo.org

Built with what exists

Small enclosed electric centrifugal blowers, such as multistage or high-pressure fan types, working below about 0.3 bar; a hand valve on the outlet as the throttle; a fast MEMS pressure sensor, a MEMS microphone and an accelerometer; a USB data-acquisition unit or a Raspberry Pi with an analog-to-digital converter; scikit-learn for the detector.

  1. First step

    Buy one small centrifugal blower, fit a valve and a pressure sensor, and record 20 slow closures to see whether a clear oscillation appears.

  2. Done looks like

    200 throttling runs on at least three blowers, each with surge onset labeled by a stated pressure-oscillation rule; on a blower it was not trained on, the detector warns at least 2 seconds before onset in 90% of runs with no more than one false alarm per hour of steady running.