Cavitation

What it is. Inside a centrifugal pump, pressure at the impeller inlet can drop below the liquid's vapor pressure. At that instant the liquid boils locally from low pressure, not from heat: every liquid boils when its pressure-temperature pair (P×T) crosses the saturation curve — ordinary kitchen boiling changes the temperature while holding pressure fixed; cavitation does the opposite, changing the pressure while holding temperature fixed. The physical result is the same — vapor bubbles form — but the trigger is the opposite one. Those bubbles, dragged into higher-pressure regions inside the impeller, collapse violently against metal surfaces.

The violence of the phenomenon. Each bubble collapsing near the wall fires a liquid microjet at 100–500 m/s — comparable to a rifle bullet — plus a local shock wave that exceeds the pump's operating pressure by thousands of times. One isolated event is harmless; the problem is that hundreds to thousands of collapses per second happen at the same spot.

How to recognize it in the field. Characteristic "pumping gravel" noise or crackling, unstable or falling flow and pressure, elevated vibration and — over time — an impeller surface pitted like an orange peel.

What cavitation costs. Three bills at once: energy (hydraulic efficiency drops, so the pump spends more to deliver less), repair (an eroded impeller demands specialized weld repair or replacement — one of the most expensive components in the assembly) and downtime (mature cavitation kills in days, not months). Plus the collateral damage: cavitation vibration accelerates mechanical seal and bearing failure — the seal commonly fails before the impeller is holed.

The underlying cause is almost always the system, not the pump: low suction-tank level, clogged strainer, or the pump running far from the point it was selected for. High fluid temperature belongs on that list too — and it is worth reinforcing the distinction above: it is not that heat "boils" the liquid directly, it is that vapor pressure rises together with temperature, eating into the pressure margin that was keeping cavitation away. Same installation, same pump: heating up the process alone can be enough to start cavitating. And the operator's golden rule: never "relieve" cavitation by closing the suction valve — that drops the inlet pressure further and makes the phenomenon worse. Reduce flow at the discharge, never at the suction.

Normative classification — the two conventions

Per ASTM G40 (wear and erosion terminology), cavitation is an erosion mode — cavitation erosion: "erosion caused by the formation and collapse of cavities in a liquid at a solid surface". ASTM G32 standardizes the material resistance test. Meanwhile ISO 14224:2016 (Table B.2) — the standard that structures this platform's taxonomy — lists cavitation as a failure mechanism in its own right, a subdivision of material failure, distinct from erosion and from corrosion. The two conventions do not conflict: ISO names the mechanism (the physical process); ASTM describes the result (erosive damage). This note uses the ISO convention's CAV code and describes the damage in ASTM terms. The distinction decides replacement material, maintenance strategy and correction.

Differential diagnosis — what cavitation is NOT

Not abrasion. Abrasive wear (Hutchings, ch. 4) requires hard particles or sliding contact, and the volume removed is governed by hardness — the harder the material, the less it wears. It removes material by tangential micro-cutting and leaves grooves oriented along the flow. In cavitation with clean fluid there is no particle and no sliding: the impact is perpendicular and produces craters with no preferred direction. Direct practical consequence: hard coatings (WC-Co) that solve abrasion have limited effectiveness against cavitation — hardness is not the governing parameter (Franc & Michel, ch. 8: hard but brittle ceramics can perform worse than softer, tougher alloys).

Not corrosion (as the primary mechanism). Corrosion pits have smooth edges, accumulated corrosion products, and appear where electrochemistry dictates (dissimilar metals, crevices, deposits). Cavitation pits have rough edges and clean metal (the impact continuously strips the passive film) and appear exactly where thermodynamics predicts — the lowest-pressure regions of the flow. The definitive proof: 316L stainless with excellent chemical resistance is severely eroded by cavitation in clean water (ASM Handbook Vol. 11).

Field diagnosis — the classic-confusions table

Four phenomena produce similar symptoms and completely different corrections. Getting this wrong means replacing an impeller to fix a flange that sucks air:

What does occur — the real mechanisms

1. Impact erosion (primary): microjet + shock wave remove material directly, forming the characteristic pits.
2. Repetitive plastic deformation with work hardening: impacts below the removal threshold deform the surface; dislocation density grows, the material work-hardens — harder and less ductile. ASTM G32 specimens show measurable surface microhardness increase before any mass loss (Franc & Michel): damage starts invisible, in the incubation period.
3. Surface fatigue: each pit edge is a stress concentrator ($K_t \approx 3$ for a circular discontinuity — Shigley, ch. 6) and the work-hardened layer has reduced fracture energy; impact cycles nucleate and propagate microcracks that coalesce and detach whole platelets — the accelerated mass-loss phase.
4. Tribocorrosion (situational): in aggressive fluids (seawater, extreme pH), impact strips the passive film and corrosion attacks bare metal before repassivation — a synergy where combined damage exceeds the sum of the parts (factors of 2–5× reported).

The macroscopic result is the autocatalytic cycle: pit → stress concentrator → microcrack → platelet loss → bigger pit. The propagation mathematics lives in the Engineer level.

Types of cavitation in pumps

• Classic suction cavitation (insufficient NPSH) — the most common; bubbles nucleate on the low-pressure face of the vanes near the impeller eye and collapse as they advance into the pressure zone. Damage on the leading edge, suction side. Worsens as flow increases (NPSHr grows with Q²).
• Suction recirculation — at low flow the fluid is no longer guided by the vanes and forms recirculation vortices at the inlet with very-low-pressure cores — even with adequate NPSHa (Fraser, 1981). Damage on the pressure side of the vanes. The higher the design's energy and suction specific speed, the earlier (at relatively higher flows) recirculation sets in.
• Discharge recirculation — operation far below BEP; backflow at the impeller exit and volute tongue. Typical signature: sub-synchronous components (0.3–0.8×RPM) and an unstable vibration level — unlike suction cavitation, which produces a stable elevated level.
• Surface vortex / insufficient submergence — low sump level drags an air+vapor rope into the nozzle (minimum submergence criteria: ANSI/HI 9.8).
• False cavitation (air entrainment) — mandatory differential diagnosis (table above): similar acoustic symptoms, completely different correction.

The "NPSH is fine but it still cavitates" case is almost always recirculation: the engineer checks the margin, finds slack and rules out cavitation — when the problem is the BEP position, not the suction.

Framework A — diagnosis

Mixed/Complex: the trigger is operational and random (it appears whenever NPSH margin vanishes or the operating point drifts from BEP), but the damage is progressive — erosion accumulates with β>1. IT-MNT-001 lists "intermittent cavitation in pumps" as the canonical example of the category — and mandates decomposition via FMEA: the correct answer to "it cavitates every time the tank runs low" is operational (level control), not a maintenance task.

Framework B — prescription

There is a detectable P-F (vibration, noise, performance trend) and the failure is evident → CBM at P-F/2 periodicity. Tasks in the exportable plan below the page. Definitive correction is operational/design — the action layers, in order of cost and reversibility:

1. Immediate operational (without stopping the machine): restore suction level/pressure; clean strainers; reduce flow at the discharge valve — never at the suction (throttling the suction drops inlet pressure and intensifies collapse; the classic operating error documented by Perez, ch. 4); reduce speed via VFD — by the similarity laws NPSHr falls with the square of speed (−20% speed ≈ −36% NPSHr, with a −49% power bonus).
2. Installation: enlarge the suction line diameter — losses fall with the fifth power of diameter (see Engineer): +20% diameter ≈ −60% losses; doubling ≈ −97%. Shorten and straighten the suction (minimum 5–10 straight diameters before the nozzle); eccentric reducer, flat side up (API 610 — a concentric one traps an air pocket); suction velocity ≤ 1.5 m/s for clean water (ANSI/HI 9.8); cool the fluid; lower the pump or raise the level.
3. Pump intervention: axial inducer at the impeller eye (cuts the set's NPSHr by 50–70% — Japikse); impeller with a larger suction eye (re-check BEP); material upgrade (a conscious palliative — buys life, does not remove the cause).

Prevention by design: NPSH margin per ANSI/HI 9.6.1 (2024) — NPSHa/NPSHr ratio of 1.1 to 2.5 by suction energy and specific speed; per API 610, NPSHa ≥ NPSHr + 1.0 m throughout the allowable operating region (AOR). Performance and NPSH testing per ISO 9906. Intake geometry per ANSI/HI 9.8.

Thermodynamics: $P_v(T)$ and the melting margin

Cavitation starts when local static pressure crosses the vapor pressure $P_v(T)$. Its temperature dependence is governed by the Clausius-Clapeyron equation (ideal vapor):

— growth is exponential, not linear. For water (Çengel & Boles, Table A-4):

The operational reading: an installation with a comfortable margin at 25 °C can cavitate severely with no physical modification at all — the process merely warming up is enough. The numerical example below quantifies exactly that. Real nucleation is heterogeneous: dissolved-gas microbubbles and cavities on particles/walls act as embryos (degassed laboratory water withstands far greater tensions before cavitating — Brennen, ch. 1).

Bubble dynamics — Rayleigh-Plesset

Growth and collapse of a spherical bubble of radius $R(t)$ is governed by:

where $P_B = P_v + P_g$ is the internal pressure (vapor + residual non-condensable gas — the term that cushions the final collapse and feeds the rebound). The left side is the liquid's inertia; $P_B - P_\infty$ is the engine: when the bubble enters the impeller's high-pressure zone, $P_\infty$ jumps, the term turns strongly negative and the collapse is unstable — the smaller $R$, the faster it accelerates. From Rayleigh's ideal case (1917) comes the collapse time of an empty cavity:

— for $R_0 = 1$ mm in water at atmospheric pressure, $\tau \approx 91$ μs: collapse is practically instantaneous, and the pressure potential energy concentrates into a volume that tends to zero.

Asymmetric collapse: microjet and shock wave

Near a wall the collapse is asymmetric (Plesset & Chapman, 1971): the bubble side facing the free fluid accelerates first, crossing the bubble as a microjet that strikes the surface perpendicularly at 100–500 m/s (visualized by Naude & Ellis, 1961). The rebound also emits a shock wave with local peaks of 1–10 GPa for microseconds within micrometre radii (Brennen, ch. 5; Franc & Michel, ch. 8).

The impact exceeds the yield strength of any engineering alloy by 1–2 orders of magnitude — over a microscopic area, for microseconds, repeatedly: at 100 to 10,000 collapses per second per cm² (Brennen, ch. 3), a cavitating pump accumulates up to ~10⁸ impacts per hour per cm² — ultra-high-cycle fatigue, far beyond the 10⁷ cycles of the conventional S-N curve. At the bubble core, sonoluminescence evidences temperatures of 5,000–15,000 K at final collapse (Suslick) — hotter than the surface of the Sun, in a nanometric volume.

From impact to mass loss — the micromechanics

The quantifiable sequence: (1) impacts above yield deform the surface plastically; work hardening raises local microhardness before any mass loss — the fingerprint of the incubation period on the G32 MDE curve. (2) Each pit edge concentrates stress ($K_t \approx 3$); the now-brittle hardened layer nucleates microcracks. (3) Propagation follows the Paris & Erdogan law:

— and since every collapse onto an existing pit raises $\Delta K$ as the crack grows, propagation accelerates non-linearly: cracks coalesce, platelets detach, and the mass-loss curve leaves incubation for maximum rate. That is why cavitation erosion is not linear in time — and why early visual inspection understates the damage.

NPSH — rigorous formulation

Applying Bernoulli from the suction reservoir level (0) to the pump inlet flange (1), with losses $\Sigma h_f$:

where $P_0$ is the absolute pressure at the liquid surface, $H_s$ the static lift (positive with the pump above the level; negative when flooded) and $\Sigma h_f$ the suction-line losses. Elevation datum: shaft centerline (horizontal pumps). Each term is an intervention lever:

• $P_0/\rho g$: fixed by altitude for an atmospheric tank — 10.33 m at sea level, ~9.2 m at 1,000 m, ~8.1 m at 2,000 m. High-plateau installations start ~1 m short of coastal ones.
• $H_s$: every metre of pump elevation subtracts a metre of NPSHa.
• $\Sigma h_f$: by Darcy-Weisbach, $h_f = f \, (L/D) \, V^2/2g$; at fixed flow, $V \propto D^{-2}$, hence $h_f \propto D^{-5}$ (f ≈ constant). The installation's strongest lever:

• $P_v(T)/\rho g$: grows exponentially with temperature (table above).

What catalog NPSHr means — and what it does not. NPSHr is NPSH₃: the value at which head has already dropped 3% in the test (ISO 9906 / HI) — a performance criterion, not a damage one. Incipient cavitation (NPSHi) occurs at NPSH well ABOVE NPSH₃ — typically 2 to 6 times, per Gülich (ch. 6): at the catalog point the pump already cavitates audibly and erodes; it merely has not yet lost 3% of head. That hierarchy is what grounds the normative margins: NPSHa/NPSHr ratio of 1.1–2.5 by suction energy (ANSI/HI 9.6.1:2024) and ≥ +1.0 m throughout the AOR (API 610).

Associated parameters: suction specific speed $N_{ss}$ (aggressive designs narrow the stable flow window and recirculate earlier); cavitation number $\sigma = (P_\infty - P_v)/(\tfrac{1}{2}\rho U^2)$, which non-dimensionalizes the flow's propensity; and the similarity law $NPSH_r \propto N^2$ — the quantitative basis of the VFD correction.

Numerical example — the same installation at two temperatures

Atmospheric tank at sea level, pump 2.0 m above minimum level, suction losses 1.2 m, NPSHr₃ = 4.0 m at the duty point.

Water at 60 °C ($\rho = 983$ kg/m³; $P_v = 19.94$ kPa): $NPSH_a = 10.51 - 2.0 - 1.2 - 2.07 = 5.24$ m → margin +1.24 m, ratio 1.31 — meets API 610 and HI for low suction energy (high energy would demand more).

The process warms to 80 °C ($\rho = 972$ kg/m³; $P_v = 47.39$ kPa): $NPSH_a = 10.63 - 2.0 - 1.2 - 4.97 = 2.46$ m → 2.46 < 4.0: outright cavitation, with nothing in the installation having changed. The options, in layer order: subcool/raise level, reduce speed (NPSHr falls with N²), lower the pump, or an inducer. Redo the whole account with your numbers — it is the suction audit that precedes any material change.

Material resistance

The correlation between hardness and cavitation resistance is weak (Hattori et al., 2004: R² < 0.5 across metallic families) — the governing parameters are fatigue limit, toughness and the capacity to work-harden without fracturing. The ASTM G32 test (20 kHz vibratory horn, 50 μm amplitude, specimen in distilled water) measures mass loss and reports MDE (mean depth of erosion) and MDER (rate). The consolidated ranking (Franc & Michel; ASM Handbook Vol. 18), as relative erosion rate (lower = better):

Grey cast iron fails so fast for a microstructural reason: graphite lamellae have near-zero tensile strength and act as a distributed pre-crack network (local $K_t$ > 5 at sharp lamella edges) — every impact finds cracks ready to propagate and connect. Intervention criteria: shallow, dispersed pits → track by boroscopy; coalescing pits or a compromised leading edge → weld repair with a resistant alloy (Co-based electrodes or high-work-hardening austenitic stainless) plus balancing; recurrence in unavoidably cavitating service → material upgrade (duplex/CA6NM) after exhausting system corrections — better material buys life, it does not remove cause.

Severity, erosion rate and the P-F

Gülich's empirical correlations (ch. 6) put the erosion rate growing with a high power of the velocity at the impeller inlet (exponent ≈ 6) and with the vapor-cavity length — severity scales brutally with speed and suction energy. That is why the P-F interval spans orders of magnitude: ~1 week (cast iron, severe cavitation, hot water) to ~6 months (stainless, moderate cavitation, light hydrocarbon) in the industrial cases of Bloch & Geitner (ch. 7). Calibrate the route for the worst plausible P-F of your material×service pair — and migrate to continuous monitoring when P-F/2 becomes shorter than the practicable route cycle.

Quantitative detection

• Vibration: broadband energy at 10–25 kHz (no dominant discrete peak), modulated by vane passing ($BPF = n_{\text{vanes}} \times N$); elevated kurtosis in the high band distinguishes cavitation's impulsiveness from other broadband noise. The stochastic signature separates cavitation from deterministic defects (bearing, misalignment); 0.3–0.8×RPM sub-synchronous points to recirculation. Baseline first: establish the healthy-machine reference at a known load condition and alarm on deviation (e.g., 2× the baseline high-band RMS) — absolute values vary machine to machine.
• Acoustic emission (100 kHz–1 MHz): captures collapse and microcrack nucleation before the conventional spectrum — the earliest stage of the P-F; demands a dedicated sensor and discipline against industrial noise.
• Pressure pulsation: dynamic transducers at suction/discharge; BPF and harmonic amplitudes grow with cavitation; low-frequency components expose recirculation instability (Gülich, ch. 10).
• Motor current (via VFD): torque oscillations from the two-phase fluid's density variation show up in the current signature — detection with no extra sensor, useful in distributed fleets.
• Performance: head drop ≥ 3% defines the test threshold; in operation, the PIMS trend of head and power against the reference curve exposes the regime before that.
• State of the art: the Liu et al. review (Processes, 2023) consolidates the field — vibration is the most widespread method, acoustic emission the earliest; ML classifiers over spectra reach high accuracies on test rigs (signal fusion outperforms single signals), with the practical bottleneck in the scarcity of labeled real-cavitation data — transfer learning is the active frontier. Hydraulic digital twins compute NPSHa in real time from process instrumentation and alarm on the margin, not the symptom.

Reliability connection — Weibull and the RCM close

Treating erosion as wear-out: β ≈ 1.5–3 over time under cavitating condition — use the calculator below for $R(t)$, MTTF and B10 sensitivity. In Fw A the category is Mixed/Complex (Nowlan & Heap: random-trigger patterns dominate): there is no characteristic failure age until the operating condition sets in — TBM is technically invalid (swapping the impeller "every N months" neither removes nor anticipates the trigger). With a detectable P-F and an evident failure, the decision is CBM at P-F/2 (Moubray, ch. 7 — predictive tasks), with continuous monitoring justified on critical or short-P-F assets. The definitive way out remains operational/design: NPSH margin and BEP position are engineering variables, not maintenance ones.