A skylight cleaning robot can work above a server room, but only with a dry-run water budget, a leak containment plan and a hard vibration limit. In 2026, most data hall operators allow robot cleaning in scheduled windows with no free water on the glass at handover.
Glass roofs over server halls are a specific problem. The room below holds equipment that costs more per square metre than the glass above it ever will. That flips every normal cleaning assumption: water is a hazard, not a tool.
Why is water the main risk over a data hall?
One drip onto a rack in the wrong place is not a cleaning problem, it is a service outage. A robot that sprays and scrubs leaves a wet film, and wet films find cracks. Older lantern skylights, especially those built before 2015 with butyl seals, are the worst offenders. We have seen seal shrinkage of 2-3 mm over a decade, which is enough to pass a slow drip under thermal cycling.
The workable method is a low-water pass: damp microfibre pads, no spray bar, and a squeegee follow-up so the glass leaves dry. On a 1,200 m2 roof that cuts water from roughly 300 L per cycle to under 40 L.
Do robots vibrate sensitive equipment?
Barely. The motion on a tracked machine is a slow traverse, not an impact. Measured at the glass surface, a robot like the Lingfeng S1 transmits far less vibration than a person walking on the roof deck in boots. The bigger transmission path is water pipes and cable trays that run from the roof slab into the hall, so route the robot away from those penetrations and you remove most of the concern.
When can you schedule the clean?
Data halls rarely have a quiet hour. The practical windows are between 02:00 and 05:00, during planned maintenance, or during a load shift when one hall is drained. If the glass sits over live racks, you need a written method statement that names the containment, the spill kit location and the person who signs off the dry check before the room is considered safe.
- Bring a leak detection pass after every cleaning cycle.
- Keep at least one dry day between cleaning and any planned maintenance below.
- Photograph the dry glass at handover, with timestamp.
Who should avoid robot cleaning over live equipment?
If the roof has open panel joints, no containment tray and a history of leaks, do not put a wet machine on it. Fix the seals first. Same if the roof slab has no drain path and standing water can migrate sideways through a cable penetration. In those cases, dry brushing with a tethered handheld tool, or accepting a dirtier roof, is the honest call.
What about certification and compliance?
Data centre clients usually ask for CE marking, an IP rating that survives a wet pass (IP65 or better on the machine body), and a documented risk assessment. Some add a client-specific permit-to-work and an ESD check if the roof is near sensitive electronics. Have those documents ready before you bid, because access approval, not cleaning quality, is what stalls these jobs. Our notes on CE and IP rules cover the paperwork side.
Is it worth doing at all?
For a 1,200 m2 roof that shields a hall running 24/7, yes, if you can do a dry handover. Dust on data centre glass raises solar gain across the whole roof, and a 4 percent gain in heat load across a hall is real cooling money. For a small 200 m2 roof over a comms room, the risk paperwork outruns the benefit. Skip it.
What does the cleaning cycle cost over a year?
For a 1,200 m2 hall roof on a monthly dry-pass cycle, the running cost is mostly operator time: roughly 8-10 hours a year of machine work plus travel and setup. Add consumables, a few hundred a year, and a low water figure because the pass uses under 40 L each time. The direct cost is small compared with the cooling benefit of keeping the glass clear.
The comparison case is manual cleaning over live racks, and that is where the numbers get uncomfortable. Access over a data hall typically means night work, two spotters, an escort for the sensitive area, and a permit-to-work for each visit. A single night clean can cost more than a whole year of robot cycles, and it carries a human-error leak risk that the robot, with a fixed damp-pad method, simply does not have.
What do operators get wrong first?
The common error is treating the data hall like any other atrium and running a normal wet pass. It works until it does not. The second is skipping the seal survey because the glass looks fine; you cannot see a 2 mm gasket crack from the roof surface, and the water test under the pane the week before is what finds it.
Third, teams forget the cable hygiene below. Hot-aisle containment, floor grilles and perforated tiles all sit under drips once a leak starts, and they channel water sideways into racks that are metres from the actual defect. Contain the path before you clean the pane, not after.
Key Takeaways
- Low-water passes cut cleaning water from about 300 L to under 40 L on a 1,200 m2 roof.
- Robot vibration is lower than foot traffic, but route away from cable and pipe penetrations.
- Realistic work windows are 02:00-05:00 or during planned maintenance.
- A timestamped dry-glass handover photo is the document that protects you.
- Roofs with open joints or standing-water paths should be sealed before any robot pass.