Orifice Flow Measurement — the Square-Root Trap

The meter that is honest at 100 % and lying at 10 %

A 4″ cooling-water line runs on an orifice plate with a 0.25 %-of-span DP transmitter. The accountant's mass balance across the unit does not close, and the discrepancy is worst on weekends — when the unit is turned down to about a tenth of its rate. The instrument technician loop-checks the transmitter at 0 % and 100 %, finds it within 0.1 %, and closes the job as "no fault found".

Guess first: the transmitter really is accurate to 0.25 %. At 10 % flow, how far out can the indicated flow be?

±12.5 %. Not 0.25 %, not 2.5 %. The transmitter was never wrong; the square root was.

The wrong belief, stated plainly: the DP signal is proportional to flow, so a 0.25 % transmitter gives a 0.25 % flow measurement. The DP signal is proportional to flow squared. Everything uncomfortable about orifice metering follows from that one exponent.

Where the square root comes from

Put a plate with a hole in a pipe. Continuity says the fluid must speed up through the hole; Bernoulli says that where it is fast, it is at low pressure. Measure the pressure difference between "before" and "at the throat" and you have measured a velocity head, not a velocity:

Continuity :  ρ·A₁·v₁ = ρ·A₂·v₂
Bernoulli  :  p₁ + ½ρv₁² = p₂ + ½ρv₂²
⇒  ΔP = ½ρ(v₂² − v₁²)          →  v ∝ √ΔP  →  Q ∝ √ΔP

Dressed in the standard's clothing (ISO 5167-2), with the real losses and the contraction rolled into a discharge coefficient C and a velocity-of-approach factor E:

q_m = (C / √(1 − β⁴)) · ε · (π/4)·d² · √(2·ΔP·ρ)          β = d/D ,  E = 1/√(1−β⁴)

C ≈ 0.60 for a sharp-edged concentric plate in fully turbulent flow (it is actually a function of β, Re_D and tapping type — the Reader-Harris/Gallagher equation). ε is the expansibility factor: exactly 1 for a liquid, and less than 1 for gas or steam, where the fluid expands as it passes the plate.

Invert it and the whole topic is in one line: ΔP ∝ Q², therefore Q ∝ √ΔP, therefore

δQ/Q = ½ · δ(ΔP)/ΔP

The square root is a friend at high flow (it halves the relative DP error) and a monster at low flow (because ΔP itself has collapsed to nothing).

Symbol key — every symbol on this sheet

Read the subscripts as words: ΔP is a difference across the tappings, q_m is mass flow, Re_D is Reynolds number built on the pipe bore, not the orifice bore.

• ΔP — differential pressure across the tappings · mbar, mmWC, kPa
• Q — volumetric flow at flowing conditions · m³/h
• q_m — mass flow · kg/s
• D — pipe internal diameter at the plate · mm
• d — orifice bore diameter · mm
• β (beta) — beta ratio = d/D, the single most consequential choice · dimensionless
• C — discharge coefficient, ≈ 0.60 for a sharp-edged plate · dimensionless
• E — velocity-of-approach factor = 1/√(1−β⁴) · dimensionless
• ε (epsilon) — expansibility factor; 1 for liquids, < 1 for gas/steam · dimensionless
• ρ (rho) — density at flowing conditions · kg/m³
• span / URV — the DP at full-scale flow; what "% of span" refers to · mbar
• ΔP_loss — permanent, unrecovered pressure loss · bar
• turndown — max ÷ min flow that still meets the accuracy target · e.g. 3:1
• Re_D — pipe Reynolds number; C drifts at low Re · dimensionless
• L_up, L_down — required straight lengths, expressed in pipe diameters D · × D

The trap, made visible

Two consequences follow directly, and both are routinely missed at the datasheet stage.

1 — Accuracy dies at low flow, so turndown is limited. Demanding ±1 % of reading from a 0.25 %-of-span transmitter gives a usable range of only about 2.8:1. That is the origin of the rule of thumb that an orifice is good for roughly 3:1 — it is not tradition, it is √(error/limit). If you need wider range, the honest fixes are a second transmitter in split range (one spanned for low flow), a smart transmitter with an of-reading spec, or a different meter — never a wishful accuracy statement on the datasheet.

2 — The square root must be extracted exactly once. Either the transmitter outputs a linearised flow signal and the DCS block is set to linear, or the transmitter outputs raw DP and the DCS does the extraction. Do it in both and the indicated flow becomes √(true): at 30 % true flow the screen reads 54.8 %, an over-read of 83 %. The vicious part is that the error is exactly zero at 0 % and at 100 % — so it survives a two-point loop check untouched and is only ever caught by a mass balance or a puzzled operator.

Slide it yourself — set pipe size, β, flow range and transmitter error, and the explorer plots the square-root DP curve against the straight line everybody imagines, then the resulting flow error across the whole range with the usable turndown marked: ▶ open the interactive: instrumentation orifice square root trap calc

Worked example — 4″ cooling water. D = 100 mm, β = 0.60 → d = 60 mm, ρ = 1000 kg/m³, full-scale 50 m³/h, transmitter 0.25 % of span.

E  = 1/√(1 − 0.60⁴) = 1.0718        A_d = (π/4)(0.060)² = 2.827e−3 m²
q_m = 1000 × 50/3600 = 13.89 kg/s
ΔP  = [q_m /(C·E·A_d)]² / (2ρ) = 29 170 Pa = 0.292 bar = 2 975 mmWC   (at 100 % flow)
At 30 % flow :  ΔP = 0.292 × 0.30² = 0.0263 bar = 268 mmWC   (9 % of span)
                0.25 % of span = 7.4 mmWC = 2.8 % of the reading
                flow error = ½ × 2.8 % = ±1.4 %
At 10 % flow :  ΔP = 30 mmWC (1 % of span) → 25 % of reading → ±12.5 %
Usable turndown for ±1 % : √(0.0025/0.02) = 0.354  →  2.8 : 1

Beta, permanent loss, and the installation that lies

Beta is the master trade-off. Small β means a small hole, a big ΔP, a strong signal and a short straight-run requirement — paid for with a large permanent pressure loss. Large β means low loss but a weak signal and long straight runs. The permanent, unrecovered loss follows ISO 5167-2:

ΔP_loss / ΔP = (√(1 − β⁴(1−C²)) − C·β²) / (√(1 − β⁴(1−C²)) + C·β²)
             ≈ 1 − β^1.9            (a good pocket approximation)

At β = 0.30 you never get 90 % of the ΔP back; at 0.60, 63 %; at 0.75, about 45 %. In the worked example the 63 % loss is 0.184 bar at 50 m³/h — about 0.26 kW of hydraulic power, roughly 400 W at the motor, some 3 400 kWh a year, for ever, to read a number. That is the real price of an orifice plate and it is almost never on the cost sheet. β = 0.4–0.6 is the usual compromise.

Straight run is not a nicety. The C in your calculation was derived on a rig with a fully developed, swirl-free profile. Put an elbow four diameters upstream and the profile arriving at the plate is skewed toward the outer wall and still rotating. C is then not the number in the calculation — a bias of several per cent that no zero-and-span calibration can see, because the transmitter is faithfully reporting a DP that no longer means what you assumed. ISO 5167-2 tabulates the required lengths against β and fitting type: for β = 0.6 a single 90° bend needs about 18 D upstream and 7 D downstream; two bends in different planes need ~48 D. Half those lengths are permitted at the price of an extra 0.5 % uncertainty, and a flow conditioner buys back some of the rest.

Watch the profile break — the 3D meter run shows the jet contracting to the vena contracta about half a diameter downstream, the pressure dipping below the downstream tap and recovering only partly, and a toggle that drops an elbow 4 D upstream so you can see the velocity profile arrive skewed: ▶ open the interactive: instrumentation orifice square root trap 3d

And yet orifices still dominate. Vortex meters lose the signal at low flow too and dislike vibration; Coriolis is superb (mass directly, wide turndown, no straight run to speak of) but is expensive and size-limited; ultrasonic and magnetic meters need clean assumptions about the fluid. The orifice survives because it is cheap, has no moving parts, works on almost any clean single-phase fluid at almost any temperature and pressure, is fully described by an international standard so two parties can agree a number without arguing, and can be recalculated on paper years later from four dimensions and a fluid property table. Know its limits and it is still the right answer surprisingly often.

Common pitfalls

• Quoting the transmitter's % of span as if it were the flow accuracy.
• Square-root extraction in both the transmitter and the DCS — invisible at 0 % and 100 %.
• Specifying 10:1 turndown on an orifice datasheet and expecting the accuracy to follow.
• Sizing β for a convenient ΔP and never costing the permanent loss over plant life.
• Taking the straight-run table as advisory. It is part of the uncertainty statement.
• Tapping orientation: liquid taps should be horizontal or below; gas taps above; steam needs condensate pots at the same level on both legs. An unequal liquid leg is a permanent DP offset.
• Installing the plate backwards. The sharp square edge must face upstream; the 45° bevel downstream. It still measures — it just measures wrongly, by several per cent.
• Forgetting that a rounded, worn edge raises the true C while the flow computer keeps using the sharp-edge value, so the meter drifts low (it under-registers). Fouling that builds up and shrinks the bore drifts it the other way.
• Using the liquid equation on gas or steam by dropping ε, and ignoring the density correction when pressure or temperature moves off the design point.

Outcome

• The orifice measures a velocity head: ΔP ∝ Q², so Q ∝ √ΔP and δQ/Q = ½·δΔP/ΔP.
• A fixed % of span error becomes a huge % of reading at low flow — 0.25 % of span is ±1.4 % of flow at 30 % and ±12.5 % at 10 %. Usable turndown for ±1 % is about 3:1.
• The square root is extracted once, in the transmitter or in the DCS, never both; doing it twice reads correctly at 0 % and 100 % and lies everywhere between.
• β trades signal strength and straight-run length against permanent loss: ΔP_loss/ΔP ≈ 1 − β^1.9, which is 63 % at β = 0.6 — real, continuous pumping power.
• Straight-run requirements are part of the uncertainty statement, not a guideline; an elbow too close biases C by several per cent invisibly.
• Interactive: ▶ open the interactive: instrumentation orifice square root trap calc — DP curve, flow error across the range, permanent loss, required straight run.
• 3D: ▶ open the interactive: instrumentation orifice square root trap 3d — vena contracta, pressure profile and recovery, tappings, elbow-too-close toggle.

Open items

• Implement the Reader-Harris/Gallagher equation for C(β, Re_D, tapping) instead of the fixed 0.60, and the gas expansibility factor ε — the gas presets are indicative only today.
• Add corner, flange and D–D/2 tapping variants and show how the downstream tap position changes the measured ΔP.
• Add a whole-life cost comparison (orifice permanent loss vs Coriolis capital) for a real duty.
• Add the low-Re branch where C departs sharply — viscous liquids on an orifice are a separate trap.

Educational teaching material — simplified and illustrative. Not for engineering design use.
© EPiC Circle · provided by PipeAxis