Can you hear that sound? Probably not, but rest assured that an artificial intelligence algorithm can.
That’s according to a new AI application that Montreal-based wholesale colocation provider Root Data Center will use to avoid the risk of suffering downtime. It will be working with San Jose-based Litbit, a company that packages connected sensors with AI in a way that simulates superhuman senses. Using this capability, Root will use AI software to listen to its power generator for potential problems.
The goal of the project is to push the data centre’s uptime even closer to 100 per cent. Root, which provides wholesale and enterprise colocation services, already operates at a reliability rating so high that a disruption to service might happen once in every 100 years. But according to Root President and CEO AJ Byers, it can do even better with AI.
“The benefit to the customer is a reduction of risk. Right now, there’s always that one in 100-year event,” he says. “Our goal is to move it to a one in 1,000-year event so it doesn’t happen in the lifetime of our customers.”
Litbit’s service trains “AI Personas” paired with machine-learning sensors. Trained by skilled data center technicians, this system can detect problems with equipment far sooner than any human could perceive a problem, and it can monitor that equipment around the clock. Litbit and Root technicians worked to train the neural network system about what the sound of its outdoor generators should be like under different conditions. For example, it might be raining outside, or windy, or hot or cold. Based on the data of this healthy sound profile, the system can compare the sound the machine is making in realtime and see if it’s outside the norm. Then it would alert technicians to investigate.
“It will listen to the generator and determine if something is off,” Byers says. “Potentially not even something that you could hear with your ear.”
Byers compares using the AI system to having 100 engineers employed full time to just stand around and vigilantly watch for problems that arise. Of course, doing that would be impossible, so Root must find another way to monitor its 50 megawatts of capacity.
A sound that could indicate a problem with the generator could range from a ball bearing that’s out of alignment, or internal fluids overheating, or any number of other mechanical issues. Root will be listening with the AI capability at 100 different points throughout its facility. While the project begins with detecting problems with the generator core, it will expand it to detect problems or improve performance monitoring of other equipment over time.
The project will also expand to include more supersenses in the future. Byers says that in the next 12-18 months, he hopes to work with Litbit on expanding the platform into video, using visual information to detect irregularities.
“It can look at the room at all times and detect something that shouldn’t be there, whether it’s a puddle or a person,” Byers says.
Root is publishing a white paper today that claims it’s the first to use machine learning to improve reliability. But Byers says that he hopes Litbit takes what they’ve learned and deploys similar solutions with other data companies. Then in the future, the industry could share data modelling techniques to identify risks.
Root operates two carrier-neutral data centres in Montreal that exceed Tier 3 standards. It announced $90 million in new financing from Goldman Sachs on Oct. 31 to invest in expansion into new regions.
