Can a Robot Inspect a Data Center Without Opening Anything?

A data center is, in the most simple terms, a building full of computers. Those computers run hot, and heat is the thing that kills them. So somebody has to walk the aisles regularly and check that nothing is overheating, nothing is blocked, and nothing has been left where it shouldn't be. That walk is called an inspection round, and at most sites a person does it with a clipboard and a handheld camera.

That’s when FacilityOps sent a robot instead.

It drives down the aisles on a set route and stops at every point it's supposed to check.

• It sees heat. A thermal camera shows temperature as color. A hot connection glows against a cool background, and the robot catches it while it's still a maintenance job rather than a fire.
• It sees the room. Cardboard boxes stacked in an equipment bay. A fire extinguisher blocked by a ladder. A door propped open that shouldn't be. These are the things that fail an audit, and a camera catches all of them.
• It sees the air. Temperature, humidity and air quality, measured at the height the computers actually pull air in, not at the one spot on a wall where someone bolted a sensor ten years ago.
• It reads the dials. Plenty of equipment has a gauge on the front and no connection to anything. The only way anyone ever reads it is by walking past and looking. Our robot walks past and looks, every time.

All of that happens without opening a single door.

Some electrical equipment sits inside sealed metal cabinets. An inspection robot can't see through steel, and neither can you.

That equipment needs a licensed specialist with the right tools to open it up and check inside, on a schedule. We don't replace that person and we're not trying to.

What we cover is everything in between. Right now, at most sites, the months between those specialist visits are blank. Nobody knows what happened. We fill that in.

Not because the person is bad at their job. Because of something more boring.

Imagine photographing the same thing every month for a year. You stand slightly closer one time. A different angle the next. Someone else covers your shift in March. Every photo is fine on its own. Line them up and they tell you nothing, because you're not comparing the same thing.

Temperature readings work exactly the same way. Move a foot closer and the number changes. So a year of readings taken by different people from different spots can't be put side by side, which means nobody can tell whether something is slowly getting worse.

The robot stops in the same place every time. Same distance, same angle, same point on the equipment. Now you can line up January and July and actually see a trend.

And it does the last stop of the route exactly as carefully as the first one, which is more than any of us manage at the end of a long shift.

This is the part we're proudest of.

If the robot can't reach a spot, or something is in the way, or the reading comes back messy, it does not make up a number. It writes down that it couldn't get the reading and why.

That sounds small. It isn't. The alternative is a record that looks complete but is incorrect, with a guess buried in it, and you find out during an audit.

It also double-checks it's looking at the right machine. Two nearly identical units sitting side by side is completely normal in a data center, and a robot confidently measuring the wrong one would be worse than useless. So it confirms the label matches the route before it records anything.

One route through the building produces one record for you. Every reading tagged with where it was taken, when, and by which pass.

And you can check that record yourself. Change anything in it and it stops matching itself, and it tells you exactly which field was changed. The checking tool runs on your own laptop and doesn't need to talk to us at all.

You don't have to take our word for any of this. That's the point.