Keeping track of the mechanical health of millions of pieces of equipment is a big job for the Army. To help with this data-intensive work, it’s recruiting an AI assistant. Machine learning software developed by Chicago firm Uptake Technologies will be used to predict when vehicles will need repair, flagging problems to army mechanics before they become disastrous.

The pilot scheme will cover a few dozen armored infantry transports (Bradley M2A3s) deployed in active service. Sensors inside the vehicles’ engines record information like temperature and RPM, and transmit this to Uptake’s software. This uses machine learning to look for patterns in the data that match known engine failures in similar vehicles.

“Our platform is like a brain that collects signals from all these nerve endings [in the engine] and produces feedback,” Uptake’s VP of communications Abby Hunt tells The Verge. “Whether it’s the coolant running low or some other problem, we know we’ve seen this in other engines, and can tell someone that the transmission may fail in a week or two, for example.”

Uptake currently provides similar services in industrial settings, and has worked with clients including Caterpillar, Boeing, and Progress Rail. For Berkshire Hathaway Energy, its software is used to keep tabs on wind turbine maintenance — an important task for expensive turbines that are often situated in remote locations.

The company’s contract with the army is relatively small, worth just $1 million, but represents a step forward for the military’s use of AI. Some experts have criticized the US for failing to take full advantage of machine learning. Others have cautioned that the Army’s internal culture is hesitant to replace human expertise with machine. Uptake stressed that their system does not itself make any decisions, but simply highlights potential problem areas for humans to investigate.

Lt. Col. Chris Conley, product manager for the US Army’s Bradley vehicles, told The Washington Post that if the trial is successful, Uptake’s software could be used to track the entire Bradley fleet and even other combat vehicles. However, Conley was cautious: “I’m not convinced that this will be successful, but I’m really excited about the potential of it,” he said. “We’re doing a pilot test to verify their claims before we do anything at scale.”