Metal is missing from an impeller, and the question is what removed it. That is not a cosmetic distinction. Classical cavitation implicates the suction system and the net positive suction head available to the machine. Suction recirculation implicates sustained operation far below design flow. Abrasive erosion implicates the process stream and whatever was meant to keep solids out of it. Erosion-corrosion implicates the fluid chemistry and the materials chosen for it. All four remove metal, and all four tend to be captioned the same way in preliminary reports.
Where the metal went matters more than how much
The volume lost indicates roughly how long the condition persisted. It does not identify the condition. Mechanism is established by where the damage sits within the machine's hydraulic pressure field and by what the surface shows under magnification.
Mapping damage onto the hydraulic geometry is the first analytical step: the impeller eye, the vane leading edge, the pressure or suction side of that vane, the shrouds, the wear rings, the volute cutwater. A report describing severe erosion without naming a face, a vane and a radius has recorded a symptom, not an argument.
Classical cavitation and what it requires
Vapour cavitation requires local static pressure to fall to the fluid's vapour pressure and then recover, so that bubbles collapse. The implosions work-harden the surface and fatigue it away. Damage sits on the low-pressure side of the vane just behind the leading edge, near the eye, and at magnification is deeply pitted, angular and spongy with no directional marks.
The condition implied is a shortfall of suction energy. Meeting the published required NPSH is not by itself a defence: the three-percent head-drop criterion is a test acceptance point, and erosion proceeds steadily at margins well above it. ANSI/HI 9.6.1 addresses margin rather than bare compliance, and API 610 and ISO 13709 treat suction conditions as a system responsibility.
Suction recirculation looks like cavitation and is not
Well below the best efficiency point, flow reverses at the impeller inlet and forms vortices whose low-pressure cores cavitate. The texture is cavitation-like, which is why the two are conflated, but the location differs: recirculation attacks the pressure side of the vane inlet and the region near the shroud, while discharge recirculation attacks the vane trailing edge instead.
The distinction decides who answers. Inlet cavitation points at suction system design, a blocked strainer or a change in fluid temperature. Recirculation points at operation outside the region the machine was sold for — the preferred and allowable operating regions and minimum continuous stable flow of ANSI/HI 9.6.3 and the API 610 datasheet.
Erosion, erosion-corrosion and chemical attack
Solid-particle erosion is directional. It burnishes and scallops the surface along the streamlines and concentrates where particles cannot follow the turn the fluid makes: the pressure side of the vane leading edge, the cutwater, the wear rings. Debris recovered from strainers and the casing corroborates it.
Erosion-corrosion leaves smooth, glossy scallops where a protective film is repeatedly stripped and reformed, and uniform chemical thinning looks different again, being largely indifferent to local velocity. Among the four, cavitation is the outlier: it alone leaves no directional signature.
Entrained gas and flashing are neither
Non-condensable gas drawn in through a suction leak or a vortexing sump produces noise, head loss and instability but little erosion, because gas bubbles do not implode the way vapour bubbles do. Flashing is the reverse case: pressure falls and does not recover, so metal loss appears downstream in valve trim and piping rather than at the impeller.
The compressor equivalents
Compressors do not cavitate, but they have their own conflated pair. Surge is a full flow reversal, producing an axial thrust excursion, thrust-bearing and labyrinth damage, and a low-frequency oscillation in discharge pressure that control data usually captures. Rotating stall is a localised cell circulating at a fraction of running speed, raising sub-synchronous vibration without reversing the machine.
Liquid carryover produces a third pattern — impeller leading-edge erosion and progressive unbalance — pointing at the suction scrubber and its level control rather than at the compressor. API 617 governs the centrifugal and axial machines and API 618 the reciprocating ones.
Confirming the mechanism in the laboratory
Scanning electron microscopy separates what the eye cannot. Cavitation pits show plastic deformation at their lips and no embedded foreign material; abrasive erosion leaves ploughing marks and often embedded particles that can themselves be analysed; corrosive attack leaves reaction products and preferential attack on a particular phase. Chemistry checked against the datasheet closes out the materials-of-construction question.
The hydraulics have to agree with the metal
A morphology finding should be corroborated by calculation. Available NPSH is reconstructed from suction vessel pressure, liquid level, line losses and the fluid temperature that actually applied, not the design case. Operating flow relative to best efficiency comes from historian data. When the damage location says recirculation and the trends show years at a fraction of best efficiency flow, two independent lines meet. When they contradict, saying so is better than resolving it by preference.
Where these opinions get tested
Predictably: that pitting was called cavitation from a site photograph without magnification; that available NPSH was taken from the datasheet rather than reconstructed at the conditions of the day; that recirculation was never separated from inlet cavitation; that solids were blamed without anyone characterising the debris.
An examination that records location, texture at magnification, the hydraulic condition each implies, and the data that corroborates or contradicts it will withstand that scrutiny. A caption on a photograph will not.
This article is general technical orientation, not a failure analysis, an engineering opinion, or advice on any specific matter. Determining the cause of a particular incident requires hands-on examination by a credentialed expert.