Never let cleaning water reach a data hall. Even a few litres onto a raised floor or through a cable tray can trigger a fault. The workable approach in 2026 is low-flow, squeegee-recovered robot cleaning, a documented method statement, and a run window aligned with the site’s change-control process.
Why is a data centre roof different?
Cooling and risk. Many halls use indirect or roof-mounted cooling, with intakes and condensers on the roof. Dirty glass is not the main concern; dirty condenser coils and blocked intakes are. Glass cleaning usually happens as part of a wider roof maintenance regime that also keeps cooling gear clear.
Then there is water discipline. Below those roofs sit servers, PDUs and cable infrastructure worth more than the building. Cleaning fluid is treated as a contaminant, so drip containment and recovery are mandatory.
What access rules apply?
Expect escort, permits and a method statement signed off in advance. No lone working. Some operators require a roof survey before every job because temporary equipment, ladders or aerials move. Battery robots avoid trailing power leads across a roof, which reduces trip and snag risk in a tight plant area.
- Permit-to-work and escort for all roof access.
- Written method statement covering spill response.
- No cleaning during cooling maintenance windows.
- Post-job roof walk to confirm no debris or pooled water.
How do you stop runoff reaching the hall?
Match water to the job. A dry or damp-microfibre pass handles light dust with almost no liquid. Where wet cleaning is needed, use a machine with squeegee recovery and cap volume per m ². Then route to drains that are not above the hall, and inspect seals and gaskets around roof penetrations before you start.
Lingkong K3 and the Lingyun Y3 both allow controlled flow. The operator’s job is to watch recovery, not to rush coverage. If a squeegee blade is worn, stop – a torn blade leaves a trail of water you will spend the rest of the day chasing.
Is a robot always the answer here?
For large, low-slope glass or roof sections, yes, because it keeps humans away from edges and cooling plant. For small, steep or fragmented glazing, rope access with a strict method statement may fit better. And if the roof has never been re-sealed, fix the sealing before adding water to the plan.
Cost is secondary to risk here. A 4,000 m ² glass roof might cost EUR 6,000-12,000 a year in rope fees. A robot at EUR 35,000-55,000 pays back in three to five years, but the honest justification is cut risk to critical equipment, not the payback.
How do you plan around cooling maintenance?
Align cleaning with the cooling calendar, not against it. If condensers are being serviced in the same week, keep cleaning out of the way and let the maintenance crew finish first. Two contractors on the same roof at once is how equipment gets knocked and intakes get blocked.
What consumables matter most here?
Squeegee blades and suction cups. A worn blade leaves water, and water is the whole risk on this site. Carry spare blades on the roof and replace at the first sign of a gap. Check cups for hardening before every run, because a release above a cable tray is the incident everyone plans to avoid.
Who signs off a data centre roof clean?
Usually facilities plus security, with the permit approved through change control. Expect a named escort and a defined window. Bring the method statement, spill kit and a contact list before anyone goes up. Sites that treat cleaning as routine access without permits are the ones that end up with an incident and a very uncomfortable review.
Does a robot reduce the need for escorts?
Not exactly, but it shortens roof time. Less time on the roof means less exposure to cooling plant, less escort resource and fewer windows blocked. On a critical site, saved hours of escorted access can be worth as much as the cleaning itself, which is easy to miss when the case is built on cleaning cost alone.
How do you handle water if there is no drain nearby?
Recover and remove it. Use a machine with high recovery and empty the tank off the roof into a safe location, never into a convenient outlet you have not confirmed. Confirm drainage routing with facilities before the first run so nobody discovers a drip tray above a comms room by accident. On critical sites, a wet-vac as backup is not overkill.
What about cleaning near air intakes?
Work when the relevant intake is isolated if the site allows it. Otherwise keep the robot and any loose dust well downwind, and clean intakes and filters as part of the roof regime. A cleaning job that loads dust into a neighbouring intake is a net negative, regardless of how clean the glass looks afterwards.
Finally, revisit the plan after any roof change. New cooling units, satellite dishes, cable runs or temporary plant alter the route and the risk. A method statement written two years ago can be quietly wrong today. Reconfirm the layout before every cleaning cycle on a critical site, and update the document when anything moves.
One final habit: brief the cleaning crew alongside the cooling team once a quarter. Two groups sharing one roof with no shared calendar is how a spill or a blocked intake eventually happens. A ten-minute joint walkthrough keeps both plans aligned and costs almost nothing.
Key Takeaways
- Treat cleaning water as a contaminant above live data halls.
- Low-flow, recovered water or dry microfibre passes reduce risk most.
- Permit, escort and a spill method statement are standard before access.
- Keep cooling intakes and condensers clear as part of the same regime.
- Justify the robot on risk reduction first, payback second.
For method statements and machine flow figures, use the contact page. Related operational notes are in the news archive.