Industries / Industrial equipment and services / Industrial machinery / Compressors
Centrifugal compressors
Designing, building and testing machines that raise gas pressure by flinging the gas outward from spinning impellers (centrifugal) or pushing it along the shaft through rows of blades (axial).
5
Milestones achieved, from 1906 to 2002.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1906
Twenty-five impellers, three casings
Brown Boveri of Baden, Switzerland, began building turbo-compressors in 1906 with Professor Auguste Rateau of Paris. The first, for the gas turbine that Armengaud and Lemale were building in Paris, took in 60 cubic meters of air a minute and had 25 impellers spread over three casings. The company's 1941 review calls it "the first multi-stage turbo-compressor in the World". Two- and three-casing machines followed, mostly to make compressed air for mines and steelworks.
LessonImpeller materials were weak at the time, the company wrote, so many stages were needed for a modest pressure.
Credit
Brown Boveri; Auguste Rateau
1932
Axial compressors go to work
Brown Boveri began making axial compressors in 1927, shaping their blades with the airfoil theory of aircraft wings, which its French associates had tested on windmills. After blowers and a wind-tunnel compressor, their main use became the Velox boiler, a boiler fired under pressure whose own exhaust turbine drives the air compressor. The first Velox boiler sold, in 1932 for the Mondeville steelworks in France, had axial compressors for the combustion air and the blast furnace gas. By 1941 about 80 axial compressors had been built for Velox boilers alone. Sun Oil used Brown Boveri's Velox compressor-and-turbine sets in its Houdry refinery units, where the turbine ran on the regenerator's exhaust and made 900 kW more than the compressor used, by Brown Boveri's account.
Credit
Brown Boveri; Sun Oil
1974
Rotors whirl at high pressure
Two gas-injection compressors of the early 1970s, known by their sites as Kaybob and Ekofisk, shook at a frequency below running speed once they met dense gas on site. This is subsynchronous whirl: a self-excited wobble of the rotor at its first natural frequency. Phillips Norway Group's two Ekofisk reinjection compressors in the North Sea were designed for 9,200 psig, almost three times what existing centrifugal gas compressors could reach, and ran successfully from December 1974 after modifications. Attempts to cure Kaybob included bearing, seal and diffuser changes, a squeeze film damper and at least two rotor redesigns.
LessonBoth cases pushed makers to analyze rotor stability before building, and led to today's stability codes and the API stability specification, Nicholas and Kocur wrote in 2005.
Credit
Phillips Norway Group (Ekofisk); case accounts by K. J. Smith and by Fowlie and Miles (Kaybob, 1975)
1984
Test at full gas density
An electric-driven centrifugal compressor built for gas reinjection on a North Sea platform could not run above about half its rated discharge pressure of 14,500 kPa (2,100 psia) when it was tested at full load with hydrocarbon gas. Shortening the bearing span, changing the flow into a central labyrinth seal and removing the vaned diffusers on all ten stages fixed it, and a retest at 105% speed proved a margin of stability. The compressor was commissioned on schedule.
LessonWhirl may appear only with gas at full density and pressure, so a test at rated gas conditions before commissioning can catch it early; Fulton also found that rule-of-thumb stability criteria correlated poorly with the cases.
Credit
J. W. Fulton, Exxon Research and Engineering
2002
Stability checks enter the standard
The seventh edition of API 617, the American Petroleum Institute's standard for axial and centrifugal compressors, added stability acceptance criteria: a quick Level I screen and a detailed Level II analysis that includes the forces from labyrinth seals, both judged by the logarithmic decrement, a measure of how fast a disturbed rotor settles. The same edition added a chapter for integrally geared compressors, in which one large bull gear drives smaller, faster pinions, each carrying impellers at its own best speed. The industrial gas industry had used such machines for more than 35 years, engineers from two industrial gas companies wrote in 2003; Atlas Copco says refineries and petrochemical plants fully accepted them after the 2002 edition.
Credit
American Petroleum Institute
Watch for
Carbon-fiber impellers for hydrogen
Hydrogen is so light that a centrifugal stage must spin its impeller tips faster than about 600 m/s to raise its pressure usefully, nearly double what natural gas needs, and API 617 bars steels with a yield strength above 827 MPa (or hardness above Rockwell C 34) where hydrogen's partial pressure exceeds 689 kPa. Greene Tweed says an impeller of carbon fiber in PEEK, a heat-resistant plastic, reached a tip speed of 688 m/s (47,800 rpm, 275 mm across) on its third design before it broke, "nearly double that of traditional metallic impellers", and that it is working with several compressor makers.
Achieved whena compressor maker runs such an impeller in a hydrogen stage and someone other than Greene Tweed reports its pressure rise and running hours.
Claimed by
Greene Tweed, February 2026
Open
01
A public whirl case table
One open table of every published case of subsynchronous whirl in a centrifugal compressor, with the machine's data (stages, speed, first critical speed, discharge pressure, gas density, bearing and seal types), whether it was unstable, and what fixed it. Then a test of how many unstable cases the API 617 Level I screen, as Nicholas and Kocur describe it, would have flagged.
The briefCredit · unclaimed02
Classic cases in open code
Rotor models of three published instability cases (Kaybob, Ekofisk and Fulton's North Sea compressor), before and after their fixes, built in an open rotordynamics code and published with every input.
The briefCredit · unclaimed
Open 01 · the brief
A public whirl case table
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
One open table of every published case of subsynchronous whirl in a centrifugal compressor, with the machine's data (stages, speed, first critical speed, discharge pressure, gas density, bearing and seal types), whether it was unstable, and what fixed it. Then a test of how many unstable cases the API 617 Level I screen, as Nicholas and Kocur describe it, would have flagged.
Why it is still open
The cases are spread over symposium papers and NASA workshop proceedings from the 1970s on. Fulton found in 1984 that empirical stability criteria correlated poorly with the cases he compared, and the experience plots in print are drawn from one maker's machines: Memmott's 2011 plots show "a representative sample of the compressors from this author's company", Dresser-Rand. Searches of Zenodo and GitHub found no compiled public set.
Built with what exists
Texas A&M Turbomachinery Symposium proceedings (open access in its OAKTrust repository, and searchable through CORE's free API at api.core.ac.uk); NASA's rotordynamic instability workshop papers on ntrs.nasa.gov; a small AI model built for extraction from long documents, such as Meta's Muse Ember (see the Note "The jobs agents repeat"); a spreadsheet.
First step
List every compressor case history in the NASA instability workshop proceedings and in the Turbomachinery Symposium case-study sessions, with page links.
Done looks like
At least 60 published cases coded, each row linked to its source page; a second person rechecks 15 rows and agrees on at least 90% of fields; a published count of the unstable cases that Level I would have passed.
Open 02 · the brief
Classic cases in open code
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Rotor models of three published instability cases (Kaybob, Ekofisk and Fulton's North Sea compressor), before and after their fixes, built in an open rotordynamics code and published with every input.
Why it is still open
The cases have been re-analyzed: Nicholas and Kocur applied API Level I and II analyses to Kaybob in 2005, and Cloud, Kocur and Pettinato revisited Ekofisk "forty years later" in 2018. Neither paper's page offers the rotor models, so no one can rerun them.
Built with what exists
ROSS, Petrobras's open-source Python library for rotordynamics (Apache 2.0), which models shafts, bearings and seals and reports the logarithmic decrement of each mode; the rotor drawings and design tables in the published papers (Nicholas and Kocur's Table 1 gives Kaybob's data); CoolProp for gas properties.
First step
Rebuild Kaybob's rotor in ROSS from Nicholas and Kocur's Table 1 and figures, and compute its first forward mode.
Done looks like
For each case, the open model gives a negative logarithmic decrement (unstable) for the original rotor and a positive one for the fixed rotor, matching the field record in all three cases; model files and a one-page comparison are published, and an outside rotordynamics engineer reviews at least one.