How to Select Valves for a Data Center Primary Loop

A primary-loop valve schedule is usually written last. The loop has been drawn, the chillers and CDUs placed, the pipe sized, and then someone opens a valve catalogue and starts assigning types by diameter: butterfly here, ball valve there, globe valve at the CDU, check valve after every pump. The schedule looks complete, and it will be wrong in three or four places that cost real money — either pump head nobody budgeted for, or a control valve that cannot hold a setpoint at part load, or a bypass that opens too slowly and trips a chiller on low flow.

The reason is that a primary loop's valves are not selected from a catalogue. They are derived from three decisions that were made before the first valve was picked: whether the loop is constant-flow or variable-flow, whether each CDU's facility-water side is being modulated or merely limited, and which way each valve must fail when the power or the signal disappears. Get those three right and the type follows almost automatically. Get them wrong and every individual valve can be correctly sized and the loop still will not run.

This article works through a primary (facility-water) loop the way the schedule should be built: architecture first, then the eight jobs a valve does, then the arithmetic that decides each type — including the two places where the conventional wisdom is backwards. All numbers use one worked example, a 2 MW facility loop at 8 K ΔT, which is roughly 60 L/s, 216 m³/h, or 950 gpm.

Decide the loop architecture before you write a single line item

Three architectures account for almost every primary loop in a liquid-cooled hall, and each produces a different valve list.

Constant-flow primary with a decoupler. Chillers see a fixed flow; a bypass (decoupler) between supply and return absorbs the difference between primary and secondary flow. The primary side needs isolation and check valves and almost nothing else, because flow through each chiller never changes. This is the easiest plant to operate and the most expensive to run.

Variable primary flow (VPF). One set of variable-speed pumps serves both the chillers and the loads. Flow through the evaporators falls with load, down to the manufacturer's minimum, below which a bypass valve opens to recirculate. VPF removes the primary pumps, the decoupler, and a set of valves, and it saves pump energy — but it converts the bypass valve and the control valves into safety-critical devices, because the chiller's minimum flow now depends on them.

CDU-limited branches on a VPF or constant-flow header. This is the architecture most AI halls actually get. Each CDU's facility side has a design flow stated on its datasheet — for example 18.8 m³/h at DN50 for a 200 kW unit and 56 m³/h at DN80 for a 600 kW unit [6] — and the CDU's own control loop modulates its secondary side, not the facility side. In that case the branch does not need a modulating valve at all. It needs a flow limiter or a balancing valve that holds the design flow regardless of header pressure. Putting a control valve there instead is a common and expensive mistake: you have added an actuator, a control loop, and a failure mode to a branch that was never meant to modulate.

Architecture	What the branch valve must do	Valve count per CDU branch	Where the energy risk sits
Constant-flow primary + decoupler	Isolate only; flow is fixed by the primary pump	2 isolation, 1 drain, 1 vent	Pump runs at full flow at all loads
Variable primary flow	Modulate to load, with a minimum-flow bypass at the plant	2 isolation, 1 control, 1 balance/limiter	Bypass valve response time vs chiller minimum flow
CDU-limited branches (facility side fixed flow)	Limit flow to the datasheet value, do not modulate	2 isolation, 1 flow limiter or balance valve	Header pressure coupling between CDUs

Table 1 — The same CDU branch needs a different third valve depending on the architecture. The valve schedule is an output of this table, not an input to it.

One more upstream decision belongs here: is the facility water treated as a fixed-flow or a fixed-ΔT resource? A CDU datasheet publishes a facility-side design flow and a design ΔT (30/38 °C, say). If the branch is allowed to modulate, the return temperature will drift at part load and the plant's ΔT collapses — which is exactly what makes chillers run inefficiently and cooling towers bleed more water. Fixed-flow branches with a flow limiter protect the plant's ΔT; modulating branches protect the branch's temperature. You cannot have both without a compensating control strategy, and the valve schedule should say which one you chose.

The eight jobs a primary-loop valve does

Every valve in the schedule exists to do one of eight jobs. Name the job first and the type becomes a short list rather than a catalogue.

Job	First choice	Do not use	Parameter that decides it
Isolate equipment for maintenance	Resilient-seated butterfly (DN80–DN600), full-port ball (DN15–DN80)	Gate valve on a line that is never cycled; butterfly as a throttle	Cycling frequency, required shutoff class, actuator torque
Modulate flow to a load	Globe or characterised ball with equal-percentage trim	Butterfly or gate for continuous throttling	Valve authority, rangeability, cavitation index
Prevent reverse flow	Dual-disc wafer check (DN50–DN300), axial check on pump discharge	Swing check on a vertical run with upward flow	Closing time, slam vs non-slam, head loss at design flow
Hold design flow in a branch	Static balance valve with metering ports, or a dynamic flow limiter	Gate valve cracked open as a "balance"	Pressure independence, setting resolution, Kvs adjustment
Protect minimum chiller flow	Modulating butterfly or globe with fail-open actuator	On/off solenoid bypass	Required bypass flow, stroke time, fail position
Relieve thermal expansion	Spring-loaded safety valve, ASME-rated	Pressure-relief "valve" from a plumbing catalogue	Set pressure vs lowest-rated component, certified capacity
Remove air and drain water	Automatic air vent at high points, ball drain at low points	Manual vent only, at one end of the loop	Loop topography, glycol concentration, vacuum resistance
Stop flow in an emergency	Actuated isolation with remote and local control	Any valve that needs a ladder to reach	Fail-safe direction, stroke time, leak class

Table 2 — Eight jobs, eight decision rules. A valve that appears in the schedule without a stated job is a valve that will be operated by whoever finds it first.

Isolation valves: stop arguing about pressure drop, start counting them

The standard argument is butterfly versus gate, and the standard evidence is pressure drop. Run the numbers on the 2 MW example and the argument mostly evaporates.

At 216 m³/h through DN200, velocity is 1.91 m/s and dynamic pressure is 1.81 kPa. A resilient-seated butterfly has a loss coefficient around 0.6 when fully open, so it drops 1.09 kPa. A full-bore gate at 0.15 drops 0.27 kPa. Expressed as flow coefficients, that is Cv ≈ 2,390 for the butterfly and Cv ≈ 4,780 for the gate. The factor of two between a resilient-seated butterfly and a full-bore gate is the number to sanity-check your own selection against: if a supplier's table claims a butterfly within 10 percent of a gate of the same size, one of the two numbers is being quoted at a different disc angle.

So the gate valve is 0.82 kPa better. On a 250 kPa pump head that is 0.3 percent. You cannot buy a pump selection, a VFD, or a control strategy on 0.3 percent, and the gate valve costs several times more, weighs four times as much, needs a crane at DN250, and — critically for a facility that runs for a decade — seizes if it is never cycled. Wedge gates that sit closed under pressure for a year develop adhesion between wedge and seat and can require destructive torque to open.

Now count them. A 2 MW loop with two chillers, two pumps, eight CDUs, and a bypass has roughly forty isolation valves at DN80 and above. Forty butterflies at 1.09 kPa add 43.6 kPa to the loop; forty gates add 10.9 kPa. The difference is 32.7 kPa, which at 60 L/s and 66 percent wire-to-water efficiency is about 3.0 kW, or 26,000 kWh a year — roughly $2,600 at $0.10/kWh.

That is a real number, and it is still not enough to justify gates everywhere: the price difference on forty DN200 valves is on the order of $80,000, which buys 26,000 kWh/year back over thirty years. What it does justify is two things. First, put the 44 kPa into the pump head budget instead of discovering it in commissioning; a 250 kPa pump becomes a 265 kPa pump, and the loop still balances. Second, spend the gate-valve money only where the valve is both large and permanent — the chiller and pump suction isolation, where a full-bore path protects NPSH and the valve is cycled twice a year.

Shutoff class deserves its own line. Resilient-seated butterflies are bubble-tight to roughly MSS SP-67 / API 598 rates; metal-seated gates are not, and a gate that weeps slightly is normal, not defective. If the isolation point is also the boundary where you plan to open a flange and work on a CDU, specify a bubble-tight seat and add a drain between the two isolation valves so the spool can be proven empty before it is opened.

Control valves: authority is a price, not a rule

The control valve is where the primary loop either behaves or hunts. Sizing starts with the valve authority:

β = ΔP_valve ÷ (ΔP_valve + ΔP_rest of circuit)

The Cv and Kv numbers used below are simplifications of the sizing equation published in ISA-75.01.01, which is the basis of IEC 60534-2-1 and of every sizing program a valve supplier will hand you [8]. Use the simplified form to check a selection, not to replace it — the standards include the piping geometry and viscosity corrections that matter once glycol is in the loop.

ASHRAE's hydronic guidance is that at full-open the control valve should drop no less than about one-third of the pump head, and the practical target is an authority of 0.3 to 0.5 [1]. The reason is not theoretical. A valve with low authority sits nearly wide open across most of its stroke, so small stem movements produce large flow changes, and the loop oscillates; below about 0.25 the valve effectively stops modulating.

But authority is bought with pump head, and pump head is bought with electricity. Here is the same 2 MW loop at five valve pressure drops, with a total circuit drop of 250 kPa:

ΔP across control valve	Authority β	Required Kv (m³/h @ 1 bar)	Throttling loss at design	Pump power at design	Cost of the extra head
30 kPa	0.12	394	1.8 kW	20.4 kW	— (unusable control)
55 kPa	0.22	291	3.3 kW	22.6 kW	baseline
80 kPa	0.32	241	4.8 kW	24.9 kW	+2.3 kW → 20,100 kWh/yr
100 kPa	0.40	216	6.0 kW	26.7 kW	+4.1 kW → 35,900 kWh/yr
125 kPa	0.50	193	7.5 kW	28.9 kW	+6.3 kW → 55,200 kWh/yr

Table 3 — Valve authority swept against pump power for a 2 MW loop at 60 L/s, 250 kPa base head, 72 percent pump efficiency and 92 percent motor efficiency. Kv = Q(m³/h) ÷ √(ΔP in bar).

Going from authority 0.22 to 0.50 costs about 6.3 kW of continuous pump power, or 55,000 kWh and $5,500 a year, and nothing in the loop gets colder. That is the price of control quality, and it is a fair price — but it should be a decision, not a default. In practice: 0.3 to 0.4 is the sweet spot for a primary loop serving CDUs, and 0.5 belongs to small terminal circuits where the pump is tiny. Note also that the throttling loss itself is not the expensive part. At 80 kPa it is 8,500 kWh a year at the actual load profile — less than the pump head penalty that produced it.

The characteristic matters more than the Cv

An equal-percentage trim with a rangeability of 50, installed at authority 0.33, delivers this installed characteristic:

Stem position	5%	10%	20%	30%	50%	70%	100%
Flow, fraction of design	0.036	0.044	0.064	0.094	0.197	0.401	1.00

Table 4 — Installed characteristic of an equal-percentage valve (R = 50) at valve authority 0.33. Half-open is 20 percent flow: that is the point of the characteristic, not a defect.

Two consequences follow. First, a control valve that is half open is passing a fifth of design flow, so "the valve is at 50 percent, the loop must be at half load" is wrong by a factor of two and a half — a mistake that shows up as a badly tuned PID loop. Second, do not oversize: select the smallest Cv that passes design flow at the calculated ΔP, because one size up moves the whole useful stroke into the top 30 percent.

Cavitation and noise set the upper ΔP limit

A control valve dropping 80 kPa on 30 °C water is nowhere near cavitation. The same valve on a 45 °C loop, at the same ΔP, is closer than the arithmetic suggests, because the allowable drop before cavitation scales with the margin between the downstream pressure and the vapour pressure of the fluid. Two practical rules cover most primary loops: keep the valve's own ΔP below about half of the upstream absolute pressure at the hottest operating condition, and never let the valve discharge into a line that can be isolated while the pump is running. A characterised ball or a globe with a hardened trim handles the margin better than a butterfly, which is one more reason butterflies should not be the default control valve on a large loop. The incipient-cavitation index and choked-flow limits behind that rule are tabulated in the standard control valve handbook, along with the noise predictions that decide whether a valve needs a low-noise cage [9].

Three-way valves: the third port is a pump

A two-way valve gives the plant a variable-flow load: flow falls with load, and the pump can slow down. A three-way valve gives the plant a constant-flow load: the pump keeps moving design flow and the surplus circulates through the bypass port. ASHRAE's hydronic chapter states the distinction plainly — the two-way valve produces a variable-flow response at the source, the three-way produces a constant-flow response, and for three-way valves the bypass port should have a linear characteristic to keep total flow uniform [1].

That bypass port is why ASHRAE 90.1 requires hydronic systems above 10 hp with modulating control valves to be designed for variable flow and to be capable of reducing pump flow to 50 percent or less, with pump demand at 50 percent flow no more than 30 percent of design wattage above 5 hp [2]. The 90.1 User's Manual is explicit that three-way valves are permitted at the end of branches only if total flow through them does not exceed 50 percent of design flow. The provision has been carried forward through every edition since [3], and the intent behind it — flow turndown, not valve type, is what the standard is buying — is spelled out in the hydronic design guidance that accompanies the standard [4].

The energy consequence on the 2 MW loop is large. A 22.6 kW primary pump running at constant flow consumes 198,000 kWh a year. The same pump on a variable-flow loop, with a realistic data hall load profile (mean load around 45 percent, weighted across seven bands) and a pump power exponent of 2.4, consumes about 71,000 kWh. The difference is 127,000 kWh, roughly $12,700 a year, from one valve-type decision made at schematic design.

Three-way valves still earn their place in three situations, and only three:

• A chiller that cannot accept variable flow. Older centrifugal machines with fixed-pass evaporators may require constant flow; the three-way valve then protects the machine, and the pump penalty is the price of not replacing it.
• Minimum-flow protection on a single-chiller plant, where a three-way bypass is simpler and more reliable than a fast-acting modulating bypass valve.
• Small systems below the 90.1 threshold, where the pump is a fraction of a kilowatt and the control simplicity is worth more than the energy.

Everywhere else, the third port is a pump running for nothing.

Check valves: design for the transient, not the steady state

A check valve is selected on its closing behaviour, and the governing case is not normal flow — it is the moment a pump stops. At 1.91 m/s in DN200 steel pipe, with a pressure-wave speed around 1,300 m/s, instantaneous stoppage produces a Joukowsky surge of:

ΔP = ρ · c · Δv = 998 × 1,300 × 1.91 ≈ 2.47 MPa ≈ 24.7 bar

The wave speed assumption of 1,300 m/s is the middle of the range for steel pipe carrying water, which spans roughly 1,200–1,480 m/s depending on wall thickness and restraint [13].

The loop is PN16. Valve class and wall thickness are set by the pressure-temperature tables in ASME B16.34 [7], and the surge above is outside every one of them at PN16. A check valve that closes in a few milliseconds — which is what a spring-loaded wafer check does — can put the pipe, the gaskets, and the CDU heat exchanger 50 percent above their rating. This is not a theoretical margin: it is the mechanism behind most of the burst gaskets and split headers that get blamed on "poor installation."

Type	Closing behaviour	Head loss at design	Where it belongs	Where it fails
Swing check	Slow, depends on reverse flow to seat	Lowest	Horizontal discharge, DN80+	Vertical up-flow; slamming on parallel pumps
Dual-disc wafer	Fast, spring-assisted	Low–medium	Pump discharge, tight spaces	Fouling in dirty loops; surge on large DN
Centre-guided / nozzle	Moderate, guided disc	Medium	High-velocity lines	Needs straight length upstream
Axial (non-slam)	Controlled, damped	Low	Large pumps, critical headers	Cost; requires correct orientation

Table 5 — Check valve selection is a transient decision. The steady-state head loss differences between these types are small compared with the surge they produce when they close.

None of the four types is a bad product; each is the wrong product for three of the four columns. The data that actually distinguishes them — closing time, seat leakage class, and Cv at the relevant disc position — is published per model, and it is worth pulling the individual datasheets rather than accepting a category recommendation from a distributor [14].

Three placement rules follow. Put a check valve on each pump discharge, never on the suction. Put one on each chiller evaporator outlet when chillers are paralleled on a common header, otherwise a stopped chiller becomes a bypass path. And on a VPF plant, put one on the bypass line so that when the bypass closes, flow cannot run backwards through it into the return header.

Where the surge risk is real — long runs, large DN, or a pump that can stop against a closed discharge — the answer is not a different check valve but a different system: a flywheel, a soft-closing actuated valve sequenced with the pump, or an accumulator. A check valve cannot protect a loop from a transient it created.

The bypass valve: the one valve that can trip the plant

On a variable primary flow plant, the bypass valve exists for exactly one reason: to keep each running chiller above its minimum evaporator flow. That minimum is not a round number chosen by the designer. It is set by evaporator tube velocity: manufacturers publish a full-load minimum corresponding to roughly 3 ft/s (0.9 m/s) of tube velocity, and the permitted minimum falls with load until it reaches a floor at about half load — in practice the floor lands near half of design flow. Field guidance puts the minimum-flow bypass capacity at 40 to 50 percent of chiller design flow [5], and published design examples for a 500 TR machine work out to 50 percent: 213 m³/h design flow, 108 m³/h minimum [12].

Size it from the worst case, which is one chiller running against a hall that has just shed its load:

• Chiller: 500 TR ≈ 1,758 kW, design ΔT 5.5 K → 276 m³/h design flow.
• Minimum flow at 49 percent (mid-band of the 40–50 percent range above) → 135 m³/h.
• Worst-case load flow → near zero, so the bypass must pass 135 m³/h.
• At a 20 kPa design drop across the bypass path, required Kv = 135 ÷ √0.20 ≈ 302 — a DN100–DN125 characterised valve, not the DN50 that "looks about right" on the drawing.

Three details decide whether that valve works when it is needed:

Fail position. Loss of power or loss of control signal should open the bypass, not close it. A closed bypass with a running chiller is a low-flow trip; an open bypass is wasted energy for a few minutes. Choose the failure you can survive.

Stroke time. The bypass must open faster than the loop's flow can fall. With a 60-second stroke and a hall whose load can drop 40 percent in ten seconds during a job abort, the valve is always behind. Specify a stroke time, and test it by triggering a load shed at commissioning.

Authority to act. The bypass should be commanded by measured chiller flow or evaporator ΔP, not by a header differential-pressure setpoint that was set once and never reset. ASHRAE 90.1 also caps the differential-pressure setpoint at 110 percent of what design flow requires and requires reset based on valve position [2]; a bypass driven by a fixed, generous ΔP setpoint quietly runs open at part load and destroys the energy case for VPF.

Relief valves and expansion: eight litres into a rigid loop

A closed primary loop is a pressure vessel whose contents expand when they warm up. Take a 2,000-litre loop, a 10 K swing, and a volumetric expansion coefficient of 0.0004 per K: the water grows by 8 litres. If there is no expansion vessel, or if the vessel is isolated behind a closed valve, that 8 litres has nowhere to go. Water's bulk modulus is about 2.2 GPa, so 8 litres into 2,000 litres — 0.4 percent volumetric strain — raises pressure by roughly 88 bar. Pipe elasticity reduces that number substantially, and it is still far above PN16.

Two valves protect this, and they are routinely underspecified:

• A diaphragm expansion vessel, sized on the loop volume and the maximum temperature, with its connection unvalved. A vessel behind an isolation valve is a vessel that will be found closed during the first hot restart.
• An ASME-rated safety valve, set at or below the lowest pressure rating of any component in the loop — not at the pump's shutoff head, and not at "a bit above operating pressure." On a PN16 loop running at 4–6 bar, a 10 bar setpoint is reasonable. On a loop with a PN10 CDU heat exchanger, it is not, and the CDU datasheet wins.

Discharge the relief valve to a safe, visible location. A relief valve that discharges onto the floor under a raised floor is a valve nobody knows has opened.

Balancing: the 14 percent that nobody measures

Three CDUs on a common header, branch runs of 20, 60, and 100 m at 200 Pa/m, each CDU heat exchanger dropping 40 kPa at design flow. Without any balancing device, the near CDU sees 44 kPa of circuit drop and the far one 60 kPa, so the far CDU gets:

√(44 ÷ 60) = 0.86 — 14 percent less flow than the near one, at identical valve position.

Fourteen percent sounds survivable until it lands on a CDU whose approach temperature is already 4 K at design: 14 percent low on facility flow raises the approach, raises the secondary supply temperature, and eats the margin that the whole thermal design was built on. The fix is not a bigger pump — it is a device that makes each branch independent of header pressure.

Device	What it holds	Best for	Watch out for
Static balance valve (manual, metering ports)	A set Kv; flow still varies with pressure	Constant-flow branches, commissioning-time balance	Needs re-balancing after any branch change
Dynamic flow limiter (pressure-independent, fixed setpoint)	Constant flow across a pressure range	CDU facility branches with a datasheet flow	Requires a minimum differential to operate; check the range
PICV (pressure-independent control valve)	Flow proportional to stem position, independent of pressure	Branches that must both modulate and stay balanced	Higher cost; CV range fixed at selection
Nothing	Whatever the header gives	Only where branches are hydraulically identical	Almost never true in a real hall

Table 6 — For CDU facility branches, the dynamic flow limiter is usually the right answer: it protects the design flow and therefore the plant's ΔT without adding a control loop.

Materials, seats and the fluid nobody designs for

Primary loops in AI halls are rarely plain water. PG25 (25 percent propylene glycol), deionised water, and corrosion-inhibited mixtures all change what the wetted parts must be.

Seats and seals. EPDM is the default for water and glycol and is incompatible with most oils; FKM resists oils and higher temperatures but swells with some glycol packages; PTFE is chemically inert and the right choice for deionised water and for anything where extractables matter. In a DI-water loop, EPDM is not automatically safe: plasticiser and filler extraction is a real mechanism, and a peroxide-cured EPDM or a PTFE seat is the defensible specification.

Bodies and trim. Bronze and brass valves are common and cheap, and they release copper ions into a DI or low-conductivity loop — which is exactly the fluid that a cold-plate loop is trying to keep clean. Above DN50, ductile iron with an epoxy coating or stainless is the usual answer; for the small-bore isolation around a CDU, 316 stainless ball valves are worth the premium because they are also the valves most likely to be opened during a maintenance window, and the small-bore product data for that duty is well documented [15].

Standardise by standard number, not by brand. The valve standards for this duty are mature and should be written into the specification by number: MSS SP-67 for resilient-seated butterflies, SP-70 for cast iron gates, SP-71 for swing checks, SP-72 and SP-110 for ball valves [10]. The hydronic piping master specifications used on public projects show how these map onto service limits and body materials [11], and they are a faster route to a defensible specification than a supplier's catalogue.

Connections. Grooved joints (AWWA C606) dominate data hall pipework because they tolerate misalignment, need no hot work, and can be taken apart in minutes — the same reason the valve bodies should be grooved or lugged rather than flanged where space is tight. Welded connections belong on the buried or inaccessible runs, and flanged connections belong where a valve will be removed with the equipment.

Put the maintenance envelope in the schedule

Half of the valves in a primary loop exist for a maintenance event that will happen exactly once — when a CDU, a pump, or a chiller is replaced. If the schedule does not describe that event, the valves will be installed in a way that makes it impossible.

• Two valves and a drain on every device. Isolation on both sides, a drain between them, and a vent at the high point of the spool. One valve and a prayer is not isolation.
• A blind flange or a removable spool where the device is a CDU on a live header. Double-block-and-bleed is acceptable for a pump; it is not acceptable for a heat exchanger you are going to cut open.
• Actuator manual override and a local position indicator on every actuated valve that can isolate flow to a load.
• Lockout points. A valve that can be closed by a technician without a lock is a valve that will be closed during someone else's maintenance.
• Air vents at every local high point and after every vertical rise, and drains at every low point. Air in a primary loop shows up as a false low-flow alarm, a noisy pump, and a CDU that cannot reach design capacity.

A 2 MW loop, scheduled

Here is the valve schedule for the worked example — two chillers, two primary pumps, eight CDU branches, VPF with a plant bypass, 250 kPa design head:

Item	Qty	Unit (USD)	Extended (USD)
DN200 butterfly, chiller isolation	8	520	4,160
DN200 butterfly, header isolation	4	520	2,080
DN200 check, pump discharge (non-slam)	4	900	3,600
DN200 butterfly, pump isolation	8	520	4,160
DN150 butterfly, bypass and drain headers	3	420	1,260
DN80 ball, CDU isolation	16	340	5,440
DN80 dynamic flow limiter, CDU branch	8	460	3,680
DN80 globe control valve + actuator (where modulation is required)	4	3,600	14,400
DN25 ball, drain and vent	24	90	2,160
DN20 ASME safety valve, 10 bar set	2	640	1,280
DN15 gauge isolation cock	12	45	540
Total, valves only			≈ 42,800

Table 7 — Approximate 2026 equipment pricing for a 2 MW facility loop, excluding actuated isolation valves, insulation and installation. About $21,000 per MW of heat rejection — roughly 2 percent of the loop's installed cost, and the cheapest insurance in the plant.

Note what the schedule does not contain: no gate valves, no three-way valves, no butterfly used as a control valve, and no balancing valve on a branch that already has a flow limiter. Each omission is a decision documented in the tables above, and each one should be defended in the design narrative — because the first value-engineering pass will try to delete the flow limiters and keep the actuators.

Ten mistakes that show up on every project

1. Putting a control valve on a fixed-flow CDU branch. You add a control loop to a branch whose CDU already controls its own secondary side. Use a limiter.
2. Sizing the control valve on line size. A DN200 line usually wants a DN100–DN150 control valve. Line-size selection guarantees an authority near 0.1.
3. Forgetting the forty isolation valves in the head budget. 44 kPa of butterfly loss is 17 percent of the pump head and is discovered at commissioning.
4. Specifying a butterfly as a throttling valve on a large header. It will erode, it will be noisy, and its characteristic is wrong for control.
5. Omitting the check valve on parallel chiller outlets and then wondering why the standby machine's evaporator warms up.
6. A bypass valve that fails closed. The one valve whose failure must be in the safe direction, specified backwards.
7. A bypass valve sized on nominal flow rather than minimum chiller flow — usually one or two sizes too small.
8. A relief valve set above the weakest component's rating, typically because the CDU's PN10 heat exchanger was not on the table when the setpoint was chosen.
9. An expansion vessel behind an isolation valve. Guaranteed to be closed at the worst possible moment.
10. No drain between isolation valves. The maintenance that was planned as a two-hour window becomes a flood and a shutdown.

Eight questions for the valve supplier

1. What is the tested shutoff class, to which standard, and can you supply the test report for the size and seat material we are buying?
2. What is the Cv/Kv at 100 percent open, and what is the rangeability of the trim you are quoting?
3. For the control valve: what authority does this selection produce in our circuit, and what is the installed characteristic at that authority?
4. What is the actuator's full-stroke time at our supply voltage, and is it adjustable?
5. What is the fail position on loss of power and on loss of signal, and how is it achieved mechanically?
6. What is the seat and seal material, and is it listed as compatible with 25 percent propylene glycol and with deionised water?
7. For the check valve: what is the closing time and the resulting surge at our flow and pipe class — and can you supply a non-slam option?
8. What is the maximum differential pressure at which this valve can be operated without cavitation damage at 45 °C?

The schedule is a record of decisions, not a parts list

Every valve in a primary loop can be traced back to one of three upstream choices: how the plant makes flow, how each branch is allowed to consume it, and what happens when the power goes away. When those three are written down, the schedule practically writes itself — butterfly here because it is cycled twice a year, limiter there because the CDU datasheet states a flow, globe at the one branch that genuinely modulates, non-slam check because the loop is PN16 and the surge is 24.7 bar.

The failures that hurt are not sizing errors. They are category errors: a throttle where an isolator belongs, a modulating valve where a limiter belongs, a bypass that fails the wrong way. Those are decisions, and decisions can be reviewed before the pipe is cut.

If you are building out a primary loop and want the valve schedule reviewed against your CDU datasheets and pump curves, send us the loop drawing and the design flow. The arithmetic above is the same arithmetic we run for the CDUs and manifolds on the other side of the heat exchanger — see our notes on setting primary and secondary water temperatures and on sizing coolant loop diameters from CDU duty.

References

1. ASHRAE. 2016 ASHRAE Handbook — HVAC Systems and Equipment, Chapter 13, Hydronic Heating and Cooling. Control valve sizing, valve authority, and two-way versus three-way load response. handbook.ashrae.org
2. ASHRAE. ANSI/ASHRAE/IES Standard 90.1, Section 6.5.4 Hydronic System Design and Control, including 6.5.4.1 Hydronic Variable Flow Systems (pump flow reduction to 50 percent, 30 percent design wattage at 50 percent flow, 110 percent differential-pressure setpoint cap). ashrae.org
3. ASHRAE. Standard 90.1 — Energy Standard for Buildings Except Low-Rise Residential Buildings, current edition overview. ashrae.org
4. ASHRAE New York Chapter. Key Impacts of ASHRAE Standards on Water Side Design, seminar handout covering 90.1 hydronic provisions, flow turndown and differential-pressure reset. ashraeny.org
5. Anvilfield. Chiller plant startup and commissioning field guide — minimum evaporator flow, minimum-flow bypass at 40–50 percent of chiller design flow, and staging proof across the load range. anvilfield.com
6. ZTMM Data Center. Floor-Mounted CDU — facility-side (primary) design flow and connection sizes, 18.8 m³/h at DN50 for 200 kW through 56 m³/h at DN80 for 600 kW. ztmmdc.com
7. ASME. B16.34 — Valves: Flanged, Threaded and Welding End, pressure-temperature ratings for valve bodies. asme.org
8. International Society of Automation. ISA-75 series — Control Valve Standards, including ISA-75.01.01 flow equations for sizing control valves. isa.org
9. Emerson. Control Valve Handbook, 5th edition — valve characteristics, rangeability, cavitation and flashing limits, actuator sizing. emerson.com
10. Manufacturers Standardization Society. MSS standard titles for valves — SP-67 Butterfly Valves, SP-70 Cast Iron Gate Valves, SP-71 Gray Iron Swing Check Valves, SP-72 Ball Valves, SP-85 Gray Iron Globe and Angle Valves, SP-110 Ball Valves. mss.org
11. Los Alamos National Laboratory. Section 23 2113 — Hydronic Piping, master specification listing the MSS valve standards and service limits referenced in this article. engstandards.lanl.gov
12. MEPVault. Variable Primary Flow Chiller Logic: Staging, Min-Flow Bypass — bypass sizing from chiller minimum flow minus load flow, and staging failure modes. mepvault.com
13. Pipeflow. Water Hammer Analysis — Joukowsky pressure surge and wave speed in liquid-filled pipe. pipeflow.com
14. Emerson. Valves and valve actuation for industrial and building services — product and selection data for isolation, control and check valves. emerson.com
15. Swagelok. Valves — small-bore stainless ball, needle and check valve selection data for instrument and sampling lines. swagelok.com