ISO 9906 Acceptance Grades: What a Passing Test Allows

A pump can pass a witnessed factory performance test and still fail to deliver the duty point you specified. That is not a loophole and it is not a vendor trick — it is exactly what the standard says should happen. ISO 9906 acceptance grades define a tolerance band around the guarantee point, and the grade you wrote into the specification decides how wide that band is. Choose it carelessly and you either pay for precision the system cannot use, or you accept a pump that is legitimately several percent short on head with no contractual recourse. Most specifications get this wrong by omission: the grade is copied forward from a previous project, nobody prices it, and the consequences surface at commissioning.

What the Guarantee Point Actually Guarantees

ISO 9906:2012 defines the guarantee point as the flow/head (Q/H) point which a tested pump shall meet, within the tolerances of the agreed acceptance class. Two things in that sentence do more work than engineers usually credit. First, the guarantee is a single point, not a curve. The manufacturer supplies a performance curve across a range of flows, but the contractual obligation attaches to one Q/H pair. Second, compliance is judged against a tolerance, so "meets the guarantee" and "produces the specified head at the specified flow" are different statements.

The geometry of that judgement matters. The tolerance is applied as a cross centred on the guaranteed point: tolerance factors are set for flow (±τQ), head (±τH) and efficiency (±τη). The head and flow guarantee is satisfied if the measured H/Q curve intersects, or at least touches, the vertical or horizontal bar of that cross. A pump can therefore satisfy the guarantee by producing the specified head at a reduced flow, or the specified flow at a reduced head — it does not have to do both.

Efficiency is evaluated differently again. The measured H/Q curve is intersected with a straight line drawn through the agreed operating point and the origin of the H/Q axes; the efficiency read at that intersection must be at least ηG(1−τη). Because efficiency is a calculated value derived from measured power input, the standards let you guarantee either a minimum efficiency or a maximum pump power input at the guarantee point — not both.

One clause routinely catches specifiers who list several operating points in a duty schedule. Additional operating points specified alongside the guarantee point are evaluated at the 3B level, regardless of the grade agreed for the guarantee point itself. Writing three duty points and a Grade 1B requirement does not buy you 1B at all three.

The Six Grades and What Each One Permits

ISO 9906 specifies three levels of acceptance across six designations: grades 1B, 1E and 1U with tighter tolerance; grades 2B and 2U with broader tolerance; and grade 3B with broader tolerance still. ANSI/HI 14.6 was harmonised with ISO 9906 and carries the same scheme. The letter tells you the shape of the band: B is bilateral, U is unilateral, and E is a special case with a unilateral efficiency tolerance on an otherwise bilateral band.

The tolerance factors applied at the guarantee point are:

• Rate of flow (τQ) — 1B: ±5%; 1E: ±5%; 1U: 0 to +10%; 2B: ±8%; 2U: 0 to +16%; 3B: ±9%
• Total head (τH) — 1B: ±3%; 1E: ±3%; 1U: 0 to +6%; 2B: ±5%; 2U: 0 to +10%; 3B: ±7%
• Pump power input (τP, optional) — 1B: +4%; 1E: +4%; 1U: +10%; 2B: +8%; 2U: +16%; 3B: +9%
• Efficiency (τη, optional) — 1B: −3%; 1E: no negative tolerance; 1U: no negative tolerance; 2B: −5%; 2U: −5%; 3B: −7%
• Below roughly 10 kW rated shaft power the default acceptance criteria widen to ±10% on flow and ±8% on head, with power or efficiency handled by a separate formula

Read the efficiency row carefully, because it is the most commonly misread figure in the table. The tolerance is a percentage of the guaranteed efficiency, not a number of percentage points. A pump guaranteed at 78% efficiency and tested to 2B may return 78 × 0.95, or about 74.1% — a loss of roughly 3.9 percentage points, not 5. Note also that there is no upper limit on efficiency; only a lower limit is established.

Why bilateral and unilateral bands change the hardware

The percentage on flow and head is the same whether the tolerance is applied bilaterally or unilaterally. What changes is where the guarantee point sits inside the acceptance box. Under a bilateral grade the guaranteed point sits at the centre of the box, and the manufacturer designs the hydraulics to land on or close to it. Under a unilateral grade the guaranteed point sits at the lower left corner of the box — but the manufacturer still designs to the centre of the box, because that is where the margin against failure is greatest.

That single geometric fact has hardware consequences. Designing to the centre of a unilateral box means selecting hydraulics for a higher flow and higher head than the duty point actually requires, which all but guarantees the pump absorbs more power than the duty demands. The manufacturer will typically fit a slightly larger impeller diameter to protect the no-negative-tolerance guarantee, and the additional power appears on the submittal curve. In one worked example published by the Hydraulic Institute, taking a selection made against a 2B market curve and re-aiming it for a 1U requirement pushed a non-overloading 100 hp curve to a non-overloading 125 hp curve.

Tolerance Bands Are Not Operating Points — the System Curve Decides

A tolerance band on the pump curve tells you almost nothing on its own. The pump runs where its curve intersects the system curve, so the same tolerance band translates into completely different operating ranges depending on system resistance. This is the part of the decision that consulting engineers are best placed to get right and most often skip.

The Hydraulic Institute publishes a worked case for a small pump with a design condition of 50 m³/h at 8.3 m and 2.5 kW pump input power. Against a flatter, more static-head-dominated system curve, a 3B band yields a flow range of 45 to 54 m³/h and a head range of 8.1 to 8.5 m, while tightening to 1B narrows that to 48 to 52 m³/h and 8.3 to 8.4 m. Against a steeper, friction-dominated system curve the 3B range narrows on its own, to 47 to 52 m³/h and 8.0 to 8.7 m.

The mechanism is straightforward once you see it. Where the system curve is steep, a vertical shift in the pump curve is absorbed largely as a change in head, because friction loss rises quickly with flow and resists the excursion. Where the system curve is flat — mostly static lift, little friction — the same vertical shift has almost nothing to push against and converts into a large change in flow. Flat system curves and parallel-pump installations are the classic cases for a tighter grade: parallel machines are easier to control when their curves are closely matched, and a flat system amplifies every percent of head error into flow error.

Specify the grade against the system curve, not the pump curve. A wide tolerance on a steep friction-dominated system may move the duty point less than a tight tolerance on a static-lift system.

The Grade You Specify Is a Manufacturing Instruction

Variation in hydraulic performance comes from manufacturing tolerances, instrument fluctuation and accuracy, driver variation, and flow instability near the suction and discharge. Narrowing the acceptance band does not make those sources disappear; it obliges the manufacturer to suppress them. Meeting the tightest and unilateral grades — particularly with efficiency or power included — can require castings with smoother surface finish and tighter control of internal dimensions and clearances, vane re-profiling, surface polishing, and special coatings. None of that is necessarily standard practice for a given part, so it adds cost and lead time. Unilateral grades may also require enhancement after an initial test, followed by a second test to confirm the enhancement worked.

Published catalogue curves are a related trap. They are normally developed to the same standards but on an early production unit, so a specific pump will differ because of impeller and casing geometry tolerances — wearing ring clearances, cast vane shapes, volute throat area, surface finish — some of which drift as patterns wear. One long-standing practitioner observation is that actual efficiencies may fall around two percentage points below catalogue values, more on some smaller standard pumps, and that this is usually recoverable through the modifications carried out during the specific pump's performance test. Impeller trim corrections via the affinity laws hold reasonably well to about 5% diameter change; beyond that the growing mismatch between impeller and casing drops efficiency and you should work from the published curve instead.

Now weigh that against what the tighter band actually buys. In the Hydraulic Institute example above, the difference between the 1B and 3B bands was at most about 3 m³/h and 0.2 m. On the power side, 3B allowed up to 2.7 kW against 2.6 kW for 1B — a 0.12 kW difference, worth on the order of €136 per year at €0.13/kWh even running continuously, against a rework that could turn a delivery measured in days into one measured in weeks. For a small pump in a non-critical service, that is a poor trade. For a large machine running continuously, the same arithmetic points the other way: restricting absorbed power through a tighter grade with a power or efficiency guarantee can be worth real money over the lifecycle, and tighter control of maximum head can lower the pressure rating required of downstream tanks and pipelines.

There is a symmetrical error worth naming. Loosening the grade also has an energy cost, because the greater allowable positive tolerance invites a larger motor and a pump that overshoots the duty point — excess head or flow that then has to be throttled or diverted, with the system running off its best efficiency point. Over-compensating for a possible negative tolerance by oversizing is the most expensive way to be safe.

Know the Defaults Before You Override Them

The standards assign default acceptance grades by application and rated shaft power, which encode a good deal of accumulated industry judgement about where precision earns its cost. For rated shaft power above 10 kW, the published defaults run broadly as follows, with the first grade applying in the 10 to 100 kW band and the second above 100 kW:

• Municipal water and wastewater — 2B, then 1B above 100 kW
• Building trades and HVAC — 2B, then 1B above 100 kW
• Electric power industry — 1B at both power levels
• Oil and gas: API pumps, pipeline and water injection — 1B
• Chemical industry, cooling tower, pulp and paper — 2B at both power levels
• Slurry — 3B at both power levels
• General industry, and dewatering, drainage and irrigation — 3B, then 2B above 100 kW

The pattern is legible: services where flow must be verifiable against a regulatory or process requirement, or where a large machine runs continuously, default tighter. Services where the pumped medium itself makes precision meaningless — slurry above all — default loose. A municipal sewage pumping station and a construction dewatering set are not the same specification problem, and the defaults say so. Departing from a default is a legitimate engineering decision; departing from it without knowing you have is not.

Acceptance Tolerance Is Not Measurement Uncertainty

These are two separate ideas that share a percentage sign, and conflating them produces bad arguments on the test floor. ISO 9906 handles measurement uncertainty and performance test acceptance grades and tolerances in separate clauses. Acceptance tolerance exists because of constructional and manufacturing variation — it is a property of the pump. Measurement uncertainty is a property of the test rig: instrument systematic error, permissible fluctuation in readings, calibration intervals, straight pipe lengths upstream and downstream of flow and pressure tappings. The standard caps permissible instrument uncertainty by grade, so the two are not meant to be traded against each other.

The care taken is finer than most people expect. For grades 2 and 3 it is sufficient to use 9.81 m/s² for gravitational acceleration; for grade 1, the standard gives a formula for the local value as a function of latitude and elevation. Results are also translated to the guarantee conditions before judgement — speed and density corrections — rather than compared raw. And note that all of this is defined for clean, cold water behaviour; NPSHR published on these tests is conventionally NPSH3, the value at which first-stage total head has already dropped by 3%, which is a cavitation-onset benchmark rather than a safe operating limit.

It follows that these tolerances stack on top of, and do not account for, performance variation introduced by real field piping at the pump inlet and outlet. A pump that lands mid-band on a well-instrumented test loop can still disappoint in a suction arrangement that the loop did not replicate.

Specifying the Test Itself

Grade is one axis; the level of verification is the other, and it is a separate commercial decision. ISO 9906 distinguishes a non-witnessed factory test, where the manufacturer collects the data and judges acceptance; a signed factory test, where the manufacturer is responsible for compliance with the agreed acceptance class and issues a signed test document; a witnessed test physically attended by the purchaser's representative, who signs off on the raw test data; and remote witnessing, where the purchaser monitors the test and the acquired data in real time from a distance.

The trade-offs are practical rather than technical. A witnessed test lets the purchaser confirm the procedure was followed and observe the machine running before shipment, but its value depends heavily on the expertise of the witness, and scheduling a witness against just-in-time manufacturing can add both cost and delay. Remote witnessing recovers most of the assurance while removing the travel and much of the schedule risk. A non-witnessed or signed test is reasonable where the purchaser already knows the model's behaviour from identical prior orders.

Where this becomes concrete is the specification you issue. State the acceptance grade explicitly rather than relying on the default; say whether efficiency or power is guaranteed, knowing you may pick one; identify which point is the guarantee point if you list several; and state the witnessing mode. AIRFIN carries out customer-witnessed hydraulic performance testing at its in-house test facility and issues material and performance certificates with the machine, covering the Weltech centrifugal pump range for industrial process, chemical transfer, dewatering, sewage handling, irrigation and high-temperature circulation duties — so the grade you specify and the rig that has to satisfy it sit under the same roof. If you are weighing a grade against its cost and lead time on a specific duty, our application engineering team can work through the trade-off with you before the specification is frozen.

The judgement to carry away is that an acceptance grade is not a quality setting to be turned up when a project feels important. It is a statement about how much the operating point is allowed to move, evaluated against a specific system curve, paid for in castings, machining time, rework and absorbed power. Work out what excursion your system can actually tolerate, check whether the default for your application already covers it, and only then decide whether the narrower band is worth the weeks and the motor frame size it will cost you. A grade chosen that way will rarely surprise anyone at commissioning.