Leak detection system commissioning field guide for data centers

Leak detection system commissioning, and what it actually proves

Leak detection system commissioning is the process of proving that a data center's water and coolant leak detection will see a leak early, tell the operator exactly where it is, and trigger the right action before the water ever reaches an energized rack. The system is a network of sensors, a controller, and the alarm and control wiring that ties it to the building. Commissioning is where you stop trusting the installer's word that it works and start applying water to it.

Water and energized electronics do not share a room politely. A data hall is full of chilled water, condensate, and now coolant loops running within inches of hardware that costs six and seven figures per rack. The whole point of the system is time. A leak caught in the first minute, at a low point, is a wet floor and a mop. The same leak caught an hour later, after it has tracked along a cable tray and dripped into a PDU, is an outage and a hardware loss.

So commissioning is not a paperwork exercise. It is the one chance to confirm, with real water, that every zone alarms, that the location readout matches the spot you wet, that the BMS gets the signal, and that the valve closes or the pump stops the way the sequence promised. Skip it and you find out whether the system works the first time there is a real leak, which is exactly the wrong time to learn.

Why does leak detection matter in a data center?

Leak detection matters because the cost of a leak in a data hall is set by how long it runs undetected, not by how much water comes out. A pinhole on a chilled water flange leaks slowly and quietly. Nobody is standing there. The hall is mostly equipment, with crews in and out, so a slow leak can run for hours under a raised floor or behind a CDU with no one to see it.

Liquid cooling has raised the stakes hard. AI and GPU racks now pull 40, 80, and past 100 kW, and the only way to move that heat is water or a water-glycol mix piped right up to the chip. There is far more liquid in the white space than there used to be, and a lot of it sits directly above or beside live electronics. The liquid cooling loop commissioning guide covers how that loop is built and flushed; this guide covers the detection layer that watches all of it, the chilled water plant included.

Early detection and fast isolation is the entire goal. Detect the leak while it is small, locate it so a tech goes straight to it instead of hunting, and isolate the source so it stops growing while someone responds. A system that alarms but cannot tell you where, or alarms but does nothing to stop the flow, has solved the easy half of the problem and left the expensive half.

What leaks in a data center?

Water gets into a data hall from more sources than most designs account for, and the detection layout has to cover all of them, not just the obvious chilled water pipe. Map the leak paths first, then place sensors on them.

The cooling units themselves are the most common offender. CRAC and CRAH units make condensate, hold a condensate pan, and carry chilled water through a coil and valve package, so they leak from the pan, the coil, the valve, and the supply and return connections. The chilled water piping is next: the mains, the branch lines, every flange, valve, strainer, and the low points where water collects. CDUs and the liquid-cooling loops add their own list, from the heat exchanger and pumps to the rack manifolds and the quick-disconnects.

Then there are the quieter sources. Humidifiers and their supply lines leak, and so does the make-up water feeding the cooling system. Roof and wall penetrations let weather in, which is why detection often follows the building envelope at the perimeter. And the leak path that actually hurts is the one that runs from any of these toward the electrical gear, the PDUs, the RPPs, and the busway, because that is where wet stops being a cleanup and starts being an outage.

Spot detectors vs sensing cable: what is the difference?

Spot detectors catch water at one point. Sensing cable catches it anywhere along its length. That is the whole distinction, and it decides where each one belongs.

A spot detector, also called a point detector, is a small sensor with two conductive contacts set near the floor. When water bridges the contacts, it alarms. It is simple, cheap, and exact about location, because the location is wherever you put it. Spot detectors are right for a defined low spot: a drip pan, a drain, the floor directly under a valve or a pump where a leak will pool. The limit is that they only see water that reaches that one point. A leak two feet away that drains somewhere else never trips it.

Sensing cable, also called leak detection rope or water-sensing cable, is a length of cable that detects water anywhere it touches along the run. You lay it along a pipe, around a unit, or across an area, and it watches the entire path, not a single point. The better controllers read the distance to the wetted spot, so the cable does not just say a leak exists, it says how far along the run it is. That is the feature that turns a hall-wide alarm into a tech walking to one flange. Most real data centers run both: cable for continuous coverage along piping and under the floor, spot detectors at the specific low points and pans where water collects.

The sensing cable and locating the leak

The value of distance-locating sensing cable is in the response time after the alarm, not before it. Any sensor can tell you a leak happened. The locating cable tells you where, in meters from the panel along the cable route, so the responder goes straight to it instead of crawling the whole underfloor with a flashlight.

How close it gets depends on the system. Manufacturers commonly quote locating accuracy within about a meter, with the addressable distance running up to 100 m or more per cable run from the panel, but the real number is the one in the product manual for the controller and cable you bought. Treat the spec sheet as the source, not a rule of thumb. The accuracy you can count on is the accuracy you verify during commissioning, not the one on the brochure.

Sensing cable is supervised, which matters more than people expect. The controller watches the cable's own continuity, so a cut cable, a disconnected connector, or a corroded splice reports as a cable fault, a separate condition from a leak. That supervision is what keeps the system honest between tests. A cable that has been pinched in a floor tile or chewed through during another trade's work tells you it is broken instead of going silently blind, and a fault alarm at 2 a.m. is annoying but far better than a dead zone you never knew about.

Where do leak sensors go in a data center?

Sensors go on the leak paths, which means under and around every water source and along every route the water would travel before it reached anything electrical. The design fails when sensors are placed where they are convenient instead of where the water actually goes.

Loop sensing cable under the front edge of each CRAC and CRAH to catch coil weeps and condensate pan overflow. Run cable along the chilled water headers, down the branch runs, and through the valve galleries, staying on the low side where a drip lands. Under the raised floor, run cable around the perimeter, across the aisle crossings, and along the pipe routes, with extra attention at joints, flanges, valves, and strainers where failures concentrate. Put cable in the drip trays and along the base of every CDU and liquid-cooling manifold. Add spot detectors at the genuine low points, the floor drains, and inside condensate pans.

Two placement rules carry most of the value. First, follow the water downhill: put sensors where a leak collects and where it would have to cross to reach the gear, not just at the source. Second, watch the path to the electrical equipment specifically. A ring of cable around a PDU lineup or a busway riser catches water tracking toward the one place you cannot afford it. Humidifiers, make-up connections, and below roof and wall penetrations round out the list.

The controller, the zones, and the location readout

The controller, often called the leak detection panel, is where every cable run and spot detector reports, and it is the piece operators actually look at when an alarm comes in. It continuously monitors each input, and when it sees water it raises a leak alarm; when it sees an open cable it raises a fault. A good panel distinguishes the two clearly, because the response to a leak and the response to a broken cable are not the same.

Zones are how the system stays readable. Rather than one giant alarm for the whole building, the cable and spot detectors are broken into zones that map to areas an operator understands: this CRAH row, that pipe gallery, this CDU. An addressable controller goes further and gives a distance along the cable within the zone, so the readout is not just zone 4 but zone 4 at 38 m, which on a mapped route is a specific flange. Multiple cable runs land on the panel as separate addressed circuits.

Size and lay out the zones during design, not after. A zone that spans half the floor saves a little wire and costs you the location benefit, because the operator is back to searching. The point of zoning is to shrink the search area to something a responder can clear in minutes.

BMS and EPMS integration, and the alarm chain

The leak panel earns its keep only if the alarm reaches a human who will act. A panel beeping in an electrical room at night, with nobody to hear it, is a sensor that detected a leak and told no one. So the panel ties into the building's monitoring layer, and that integration is part of the commissioned system, not an optional extra.

Locally, the panel gives an audible and visual alarm at the unit. Beyond that, it reports to the building management system, and on larger sites to the DCIM platform and the electrical power monitoring system, so the leak shows up on the operator's screen and at the network operations center alongside everything else they watch. The BMS topic, covered in the building automation material, is where the alarm gets routed, escalated, and turned into a notification to whoever is on call. The mechanical commissioning of the chilled water and cooling plant ties into the same monitoring layer.

Design the alarm chain explicitly. Who gets notified, what the first action is, and when an automatic shutdown is allowed to fire, all of that has to be defined before commissioning, because the test confirms the chain, and you cannot test a chain that was never specified. A leak alarm that lands on a screen nobody is assigned to watch is the same as no alarm at all.

Automatic isolation: the interlock that stops the leak

Detection tells you there is a leak. Isolation stops it from getting worse while someone responds, and on a fast-leaking loop the difference between the two is the difference between a wet floor and a flooded row. The interlock is where the leak alarm drives an action instead of only an annunciation.

On a confirmed leak in a given zone, the system can close a motorized or solenoid isolation valve feeding that loop segment, shut down a CDU, or stop a pump, cutting the source so the leak cannot keep feeding. Some setups run the isolation through the BMS, where the leak signal triggers the valve and pump commands; others wire a hard interlock from the leak panel straight to the valve for speed. Both are legitimate. The choice depends on how fast the action has to happen and how much you trust the path.

Be deliberate about which leaks get an automatic action and which only alarm. Slamming a chilled water valve shut takes cooling away from a hall full of running equipment, so an automatic isolation that misfires on a nuisance alarm trades a small wet spot for a thermal event. The usual answer is to reserve automatic isolation for the loops where a leak is most dangerous and the action is most contained, like a single rack manifold or a CDU, and to leave the larger plant isolation as an operator decision off the alarm. Whatever the design picks, commissioning has to prove the action fires correctly and only when it should.

Installing the sensing cable

The cable only works where it makes contact with a real leak path and stays clean enough to sense water. A run that is laid wrong nuisance-alarms, misses leaks, or both, and no amount of commissioning fixes a bad install. It only finds it.

Route the sensing cable along the pipe or across the floor area it protects and clip it down so it stays put and keeps contact with the surface where water will land. Keep it off contaminants. Oil, conductive dust, and standing cleaning chemicals can bridge the cable the way water does and produce a false leak, so the cable does not belong in a spot that is routinely wet or dirty for reasons that have nothing to do with a leak. Use leader cable, the non-sensing jumper, to span dry stretches and to connect the sensing runs back to the panel, so the sensing portion only covers areas you actually want watched.

The classic install mistake is running sensing cable right where normal condensate falls, under a coil that always drips a little, or across a drain that sees water as part of doing its job. That cable will alarm on normal operation, the operators will learn to ignore it, and then it cries wolf the day the leak is real. Keep the sensing cable on the leak paths and off the places that are wet by design.

Why does my leak detection keep false-alarming?

Nuisance alarms almost always trace to placement and contamination, not to a bad sensor. The cable or spot detector is sitting somewhere that gets wet or dirty as part of normal operation, so it reports water that is not a leak. Fix the location before you touch the sensitivity.

The usual culprits: sensing cable run under a coil that always sheds a little condensate, cable laid across or too near a floor drain, a spot detector in a pan that holds a normal film of water, or cable fouled by oil, conductive dust, or leftover cleaning chemical that bridges it like water. Sometimes it is mechanical, a pinched or damaged cable reading as a fault, which a good panel reports as a cable fault rather than a leak if you read the panel instead of just hearing the beep.

Sensitivity has a place, but it is the second lever, not the first. Most systems let you set a threshold or a time delay so a momentary trace does not trip an alarm, which is reasonable tuning. What is not reasonable is cranking the sensitivity down to silence a cable that is in the wrong place, because now you have a sensor that ignores the real leak too. Move the cable off the normally wet area, clean it, repair the damaged section, then tune. A system that nuisance-alarms gets ignored, and an ignored leak detection system is worse than none, because everyone believes they are covered.

Spot detectors and the condensate drain pans

The condensate pan is where a CRAC or CRAH leak shows up first, so it gets its own dedicated sensor. The pan catches coil condensate during normal cooling and routes it to a drain. When the drain clogs or the float fails, the pan overflows, and that overflow lands on the floor under the unit and heads for whatever is downhill.

A spot detector or a pan float switch in the condensate pan catches the rising water before it ever reaches the lip. This is a different signal from a leak on the chilled water side, and treating it as its own point pays off, because a clogged condensate drain is a slow, recurring nuisance that you want flagged early and specifically, not lumped into a general leak zone.

Pair the pan sensor with cable looped under the unit. The pan float catches the overflow at the source; the cable under the front edge catches a coil weep, a valve drip, or a connection leak that never makes it to the pan. Between the two, a wet CRAC announces itself wherever the water actually comes from.

How do you commission and test a leak detection system?

You commission a leak detection system by applying real water to every zone and confirming the system does all four of its jobs: detects the water, locates it, notifies the BMS, and triggers the isolation the sequence calls for. This simulated leak test is the heart of the commissioning, and there is no shortcut for it. A visual inspection that the cable is installed does not prove it senses water, locates correctly, or talks to the BMS. Only water proves that.

Work it zone by zone, every zone, with nothing skipped. For each one, apply a small amount of clean water to the sensing cable or spot detector, then go to the panel and confirm the right zone alarms. Confirm the location readout points to where you actually wet the cable, not to a different spot. Confirm the alarm reaches the BMS and the operator screen, and that the notification fires to whoever the response plan names. Where that zone is tied to an automatic action, confirm the valve closes, the CDU shuts, or the pump stops, and confirm it does so for that zone and not for a neighbor. Then dry the sensor, confirm the system clears and resets, and move to the next zone.

Test the location accuracy as part of this, not as an afterthought. Wet the cable at a known point, read the distance the panel reports, and check it against where you stood. If the readout says 40 m and the water is at 25 m, the cable map or the addressing is wrong, and an operator chasing a real leak will go to the wrong place. The accuracy you accept is the accuracy you measured here, against the manufacturer's stated tolerance. Document every zone: which sensor, where, did it alarm, did it locate correctly, did the BMS receive it, did the isolation fire. A zone that was not water-tested was not commissioned.

The integration test: panel, BMS, valve, and the response

The simulated leak test proves a zone end to end, and the integration test is the part of it that proves the whole chain hangs together instead of just the sensor. A panel that alarms locally but never reaches the BMS, or a BMS that gets the alarm but never closes the valve, is a broken chain that each trade can swear is fine on their own piece.

Run the chain as one event. Apply water, watch the panel alarm, confirm the BMS and DCIM receive the point with the correct zone and description, confirm the notification goes out to the on-call path, and confirm the interlocked valve or pump acts. Then confirm the reset: the valve reopens or the pump restarts under operator control, the alarm clears at the panel and at the BMS, and the system returns to normal. Each handoff between systems is a place the integration breaks, and the test exists to find those handoffs before a real leak does.

Fold this into the overall commissioning, not as a standalone island. Leak detection is one system among the mechanical, electrical, and fire systems that all report to the same monitoring layer and all get integration-tested together. The fire and life-safety guide covers the cause-and-effect testing that proves those interlocks; leak detection belongs in the same integrated test mindset, where the question is always whether the signal makes it all the way to the action.

Mapping the cable to the floor plan

A location alarm is only as useful as the map that turns meters into a place. When the panel says zone 4 at 38 m, the operator needs an as-built that shows where the cable for zone 4 runs and what sits at 38 m along it. Without that map, the distance readout is a number with no meaning, and the responder is back to searching.

Produce an accurate as-built of the cable route and the zone boundaries, marked on the floor plan, showing where each run starts, how it routes, and what equipment it passes. The map should let someone reading a location alarm walk straight to the spot. This is not a design drawing of intent; it is a record of where the cable actually went in, because cable gets re-routed during install around obstructions the design never saw.

Hand the map to the operations team as part of the turnover package and confirm it against the location accuracy test. If commissioning wet the cable at a known point and the panel reported a distance, the map has to put that distance at that point. When they disagree, the map is wrong or the addressing is, and you fix it before turnover, not during the first real leak.

Leak detection for liquid cooling and the dielectric question

Liquid cooling moves the leak risk inside the rack, right onto the hardware, so detection moves in with it. The chilled water plant is still out in the gallery, but the direct-to-chip loop runs coolant up a manifold and through cold plates clamped to the chips, and a leak there lands on the most expensive part of the room. In-rack and at-manifold leak detection, drip trays with sensors under the CDU and the manifolds, and detection at the quick-disconnects are how the liquid loop gets watched.

The fluid changes the sensor choice. Standard sensing cable and spot detectors key on conductive water, which is fine for chilled water and a water-glycol mix. A dielectric coolant, the kind used so a leak onto electronics does not short them, is non-conductive, so a conductivity-based sensor will not see it. Detecting a dielectric leak needs a sensor suited to that fluid, an optical or capacitance type or a float, depending on the system. Confirm the sensor technology matches the actual coolant, because a water sensor watching a dielectric loop is decoration.

Immersion cooling is its own case again. The hardware sits in a bath of dielectric fluid, so the question is not a drip on the floor but the fluid level and the integrity of the tank, which is a level and containment problem more than a drip-detection one. The liquid cooling loop commissioning guide covers the loop construction, the flush, and the hydrotest that prove the loop leak-tight in the first place; this detection layer is the backstop that watches it in service after the GPUs are in.

The maintenance the owner takes on

Leak detection is not install-and-forget, and the owner takes on a real maintenance burden the day the project turns over. Sensing cable lives on the floor and under the floor, exactly where it collects dust, gets stepped on during other work, and gets re-routed by whoever is in the underfloor next. Left alone, a system that passed commissioning quietly degrades.

The recurring tasks are straightforward but they have to actually happen. Re-test the zones with water on a periodic schedule, the same simulated leak test from commissioning, because a sensor that worked at turnover may have been fouled, damaged, or disconnected since. Keep the sensing cable clean and clear of contaminants and standing debris, since the same oil and dust that cause nuisance alarms also blind the cable to real water. Check the panel and its backup battery, because a leak panel that loses power during an outage, the exact moment cooling equipment is cycling and most likely to fail, is watching nothing.

Fold the periodic test into the site's overall inspection, testing, and maintenance program alongside the fire and mechanical systems. The owner who treats leak detection as a one-time commissioning item, never tested again, has a system that was proven once and is now an unknown. The whole value is that it works on the day of the leak, and the only way to know it will is to keep testing it.

Field checklist

What to document

The commissioning record is what proves the system was tested and what the operator uses to act on an alarm later. Skip the zone-by-zone test record and there is no evidence any zone was ever wetted and proven, so a real alarm a year out rests on nothing. Capture the test results per zone and the map together, so the location alarm a year from now points to a place someone can find.

For each zone, record the sensor type and location, whether the simulated leak test alarmed, whether the location readout matched the actual spot and by how much, whether the BMS received the alarm, and whether the automatic isolation fired correctly. Tie the whole set to the as-built cable map. Log the field record where the operations team will actually find it; FieldOS keeps the per-zone test results and the turnover package in one place instead of a binder that gets lost. The test you cannot show is the test that did not happen, as far as the next owner is concerned.

Common mistakes

• Placing sensors where they are convenient instead of on the real leak paths, so the water never reaches a sensor before it reaches the gear.
• Running sensing cable under a normally dripping coil or across a drain, so it nuisance-alarms until everyone ignores it.
• Commissioning by inspection instead of with water, leaving zones that were never proven to actually detect a leak.
• Skipping the location accuracy check, so the distance readout points an operator to the wrong spot during a real leak.
• Turning over no cable map, leaving a location alarm that is a number with no place attached.
• Stopping at the local alarm, with no proven path to the BMS, the notification, or any operator who will respond.
• Detecting but never acting: no interlock to close a valve, shut a CDU, or stop a pump on a fast-leaking loop.
• Watching a dielectric coolant loop with a conductivity-only water sensor that cannot see the fluid.
• Treating commissioning as one-time, with no periodic re-test, so the system silently degrades after turnover.

Standards and references

The leak detection manufacturer's documentation is the controlling reference for the system itself. The cable and panel specifications, the locating accuracy, the addressable distance per run, the installation requirements, and the test procedure all come from the manufacturer, and that is the document you commission against. When this guide gives a number like roughly a meter of locating accuracy or up to 100 m per run, treat it as typical and confirm the actual figure for your equipment in the product manual.

The surrounding standards set the context the detection lives in. Data center infrastructure guidance such as TIA-942 and the Uptime Institute Tier framework speaks to redundancy and the reliability of mission-critical facilities, which is why leak detection exists at all. ASHRAE TC 9.9 thermal guidelines and liquid-cooling material govern the cooling design the detection protects, and the supply temperatures and coolant the loop runs. The commissioning process itself follows the project's commissioning plan, with ASHRAE Guideline 0 commonly cited as the framework for the commissioning process across building systems.

Above all of it sits the project specification. Where the spec, the manufacturer, and a general standard disagree, the contract documents control, and the editions adopted for the project govern over any general reference. Cite the standard that actually controls the point, confirm the section against the adopted edition before you put it on a submittal, and let the project spec override a rule of thumb whenever it is stricter.

Units, terms, and conversions

Leak detection carries a few names and a couple of unit systems, so the same parts read differently across a manufacturer sheet, a spec, and a drawing set.

Sensing cable is also called leak detection rope or water-sensing cable. A spot detector is also called a point detector. Distance along a cable is given in meters in most manufacturer documentation, since the systems are commonly specified in metric, while the building drawings may be in feet, so 100 m is about 328 ft and a meter of locating accuracy is about 3.3 ft. The panel may be called a leak detection controller, and the alarm layer it reports to is the BMS, with DCIM and the EPMS as the related monitoring platforms on larger sites.

Sensing cable

Leak detection rope that detects water anywhere along its length and, on addressable systems, reports the distance to the leak

Spot detector

Point sensor that detects water at one fixed location, used in pans, at drains, and at low points

Leader cable

Non-sensing jumper cable that spans dry areas and connects sensing runs back to the panel

Zone

A mapped area or cable circuit that alarms as a unit, sized small so a responder can find the leak fast

Cable fault

A break or disconnection in the sensing cable, reported separately from a leak by a supervised system

Interlock / isolation

An automatic action on a leak alarm that closes a valve, shuts a CDU, or stops a pump to cut the source

CDU

Coolant distribution unit, the heat exchanger and pump skid between the facility water and the rack coolant loop

Dielectric coolant

Non-conductive cooling fluid that a conductivity-based water sensor cannot detect, requiring a fluid-matched sensor

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What is direct-to-chip liquid cooling?
• What fire suppression do data centers use?
• Why do data centers need liquid-cooling leak detection?
• What is a CDU in a data center?
• What is data center commissioning?