Destroyer Has Become First U.S. Navy Ship To Deploy Artificial Intelligence System

The Navy destroyer USS Fitzgerald (DDG-62) became the first warship to deploy with a program-of-record artificial intelligence (AI) platform. Its creators say the system will help the fleet predict and tackle maintenance needs in a far less disruptive fashion. The system aims to reduce surprise equipment casualties while ensuring that more of the fleet is available should an all-out war break out, requiring a surge of forces.

Known as Enterprise Remote Monitoring Version 4 (ERM v4), the system is the shipboard aspect of a Pentagon program called Condition Based Maintenance Plus, which in part aims to leverage machine learning to help ship crews, ashore commands, logistical nodes, and other units keep more assets ready to fight, Zac Staples, a retired Navy officer whose Austin-based company, Fathom5, created the system, told TWZ Wednesday. Staples spoke at the annual WEST conference in San Diego this week, which TWZ attended, about his company’s innovation before chatting with TWZ.

“Right now, you’ve got about a third of the Navy that is deployed, a third of the Navy that’s in some depot-level maintenance, very difficult for them to get put back together in surge, but that other third is in various stages of material and training readiness to deploy,” he said. “Anything, particularly AI, that tells us exactly what we need to do to keep a ship ready before its readiness dips directly contributes to having a much larger battle fleet ready to defend the interests of the nation.”

The AI system arrives in the fleet as the Navy continues to suffer severe maintenance delays and readiness issues, even as it races to prepare for a potential war with China. And while the sea service is experimenting with AI in a large variety of ways, Staples said ERM v4 is the first program-of-record platform deployed in such a fashion aboard a Navy ship. It intends to replace the legacy Integrated Condition Assessment System (ICAS), which has been in use since the 1990s, according to the Navy.

On the Arleigh Burke class USS Fitzgerald, ERM v4 evaluated about 10,000 sensor readings per second coming from the ship’s hull, mechanical and electrical (HME) systems, with AI algorithms making maintenance recommendations that directly feed into the ship’s maintenance planning system, he said.

In one instance, ERM v4’s analysis identified a so-called “long lead item,” or ship part, that was nearing failure and would take a long time to receive after ordering, Staples said. ERM v4 alerted the crew to the part’s degradation far enough in advance for them to order a replacement and pick it up at the pier, “to make a repair that, otherwise, probably would have put a particular system out of commission for a bit.”

Staples declined to identify the part, deferring to Naval Sea Systems Command (NAVSEA), which did not respond to TWZ questions by deadline Wednesday. But according to Staples, ERM v4 gave sailors visibility on an impending failure that they wouldn’t have been aware of  otherwise.

“A lot of systems on the ship are operated until failure, then repair or replace,” he said. “This gave the crew, and then the shoreside maintenance community, visibility of an impending failure, so that the reliability stayed higher.”

Staples likened the evolution of Navy maintenance from an interval to a condition-based system, to advances in car oil changes. Previously, drivers were told to just change their oil every 3,000 miles. But newer models monitor mileage, oil viscosity and temperature, and alert the driver when the oil needs to be changed.

Navy officials have acknowledged that taking on some precautionary maintenance, instead of waiting for scheduled maintenance times to analyze a ship and make repairs, comprises a sea change in how the service has traditionally handled such work.

“There are numerous challenges, both internally and externally, in front of us as a program,” Mathias Haegele, a mechanical engineer with the Naval Surface Warfare Center, Philadelphia Division, said in 2023 when discussing condition based maintenance. “Some of the new and innovative products that we are developing and preparing to deploy are going to require a change in culture and how we interact with our end users to encourage engagement, adoption, and continuous feedback.”

The system takes up about half a server rack on a ship, Staples said, and it’s changing how sailors read and report system data as well.

“There are still a bunch of systems on the ship that are manually read,” Staples said. “Today, on most ships, you’ve got sailors going around with clipboards, and they’re writing down pressures and temperatures from manual gauges.”

But those readings are critical to feeding AI algorithms accurate data to grow and refine the analytics, so ERM v4 has also involved giving sailors smartphone-like devices to take logs digitally.

“Turns out, they’re much faster at a texting type operation,” Staples said. “The real difference that a lot of people notice is they shift over to digital log keeping for several of the rounds.”

All of ERM v4’s maintenance recommendations are back-checked by shipboard maintenance leadership that confirms those recommendations, Staples said, and Fathom5 and the Naval Surface Warfare Center Philadelphia, the oversight entity for the effort, take in that feedback, which is used to refine the system’s analytic potency.

“That feedback from sailors about the accuracy and relevancy of the recommendations is absolutely vital to this [machine learning] ops vision, where AI is actually reliable and trustable on maintaining things,” he said.

ERM v4 is also updated four times a year, which helps create a machine learning loop where models are trained and developed in the cloud, deployed aboard a ship, measured for performance and accuracy, and then tuned and redeployed “in a very quick succession,” Staples said.

By comparison, the Submarine Warfare Federated Tactical System (SWFTS), considered the gold standard of continuous update and delivery, is on a two-year cycle, he said.

While ERM v4’s initial use has been in ship engineering maintenance, such condition-based maintenance would be applicable in the future to combat systems, although Staples acknowledged that that would be “a more complicated place to start the program.”

Staples said that ERM v4 will be integrated into the Naval Maintenance Repair and Overhaul (NMRO) system this year, becoming the predictive maintenance and AI layer for that larger logistics and IT portfolio. The system will also deploy four more times in 2025, feeding the AI analytics more data on which to make more refined recommendations.

The capability will be getting onto one or two more ships this year, including potentially aboard an amphibious transport dock, he said.

“Then we should see a quick scale, starting in 2026, with a dozen or more ships a year,” Staples said.

The Navy has for years wanted to better capture and crunch a ship’s HME data in order to make maintenance less disruptive to a vessel’s operations and general readiness. An earlier ERM system was tested back in 2019, according to a USNI News report, which cited now-retired Rear Adm. Lorin Selby, then NAVSEA’s chief engineer and a proponent of the capability.

USNI News noted how previous, larger-scale CBM efforts had led to maintenance availability disruptions:

“During previous attempts at incorporating CBM, there was a thought that, if major efforts like refurbishing tanks were only done when needed, rather than on a predetermined timetable, the Navy could avoid spending time and money on work ahead of need. However, that also meant that shipyards wouldn’t have a clear work package before a ship showed up at the pier, adding uncertainty and, ultimately, more time and cost into the maintenance availability.”

“This time around, Selby sees condition-based maintenance as a way to address smaller maintenance items in such a way that data analysis points a ship crew to components that are experiencing minor performance issues or otherwise showing signs they are about to fail before the failure actually occurs.”

While some machine learning has already existed on Navy surface combatants to a lesser degree, particularly within its ever-evolving Aegis Combat System, using a true AI system that was designed from the ground-up has major benefits. In this case, for keeping ships out to sea and on station.

Aside from helping crews spot failing parts faster, a honed AI system for maintenance could also help with logistics and the pre-positioning of certain components before a future degradation is detected. Fleet-wide, it could generally know what parts are more prone to degrade, and when, across numerous classes. It remains unclear how clearly the Navy has captured such data and acted upon it in the pre-AI era.

Using real time health monitoring and automation to forecast what parts will be needed, and when, has also been a major feature — for better or worse — of the stealthy F-35 Joint Strike Fighter program, as well as other new military weapon systems and is becoming a critical requirement for future platforms. The commercial space has also been very bullish on the concept for all types of vehicles and mechanical systems. Enhanced AI agents are thought to only increase just how effective these systems are and how much efficiency can be gained in complex logistics chains that support all kinds of systems.

AI and machine learning are moving at a rapid pace, transforming various sectors of society. Having such a system constantly on watch to flag future ship system failures could go a long way to ensuring that Navy ships can carry out their missions around the world and not have their cruise schedule sidelined by a faulty widget.

Contact the author: [email protected]