CHICAGO — Artificial intelligence is weaving itself into the daily operations of big U.S. railroads, helping managers make data-backed decisions about operating plans and identify trends among virtually any process.
As railroads migrate from mainframe-based systems to cloud-based platforms with real-time data components, railroads are plugging in generative AI — machine-learning tools — to make data-supported decisions. There are dozens of ways to slice the data railroads are capturing from the emerging technology.
Union Pacific is one of the most vocal proponents of the tech, frequently discussing AI on its online blog, Inside Track, and at industry events. UP recently went in depth with Trains News Wire on how AI is being used at North America’s largest railroad.
“In 2024, Union Pacific completed a significant chapter of operating systems modernization for three operating platforms: Positive Train Control (PTC), Computer Aided Dispatch (CADx) and Net Control, our TMS transportation management system,” says spokeswoman Robynn Tysver. “We are the first Class I railroad to modernize the ‘big three’ operating platforms.”
The integration of these three systems has provided the railroad with rich, real-time information simultaneously flowing from hundreds of thousands of activities across its network. Combined with machine-learning AI, this provides the railroad’s operating team with insights into how to use people and equipment more efficiently, or schedule trains at certain times to maximize fluidity across the network. AI can evaluate and provide recommendations into virtually any business process.
UP discussed three main areas it uses AI today.
First, through transportation planning, the railroad uses AI to develop plans that meet changing demands for its operating teams. For example, it is being used to create models that can ascertain network resource levels, rail traffic demand and patterns, and balance all of those variables to develop the most optimal operating plans. Evaluating all types of operational scenarios, AI makes its best recommendation based on current circumstances.
Second, the railroad is tapping AI when it needs to pivot from an established operating plan to a new plan because of an event like a derailment or weather. “We operate a large outdoor factory, and things change rapidly,” says Tysver.
Third, the railroad has developed a new terminal command center that serves as a dashboard for its operating team who manages rail terminals. The center uses technology to recognize all actions of the railroad’s frontline employees and calculates the cascading effects of those decisions. In the future, the railroad plans on harnessing AI within the center to develop turn-by-turn directions that help drive increased consistency.
In other words, the railroad sees an opportunity to use AI to study the decisions of railroaders and subsequent train movements across terminals, over a period of time, to develop the most efficient operating practices within those terminals.
These resources not only help the railroad’s operating team make better decisions, but this data is also communicated to provide shippers with better information.
The railroad rolled out its new customer vision mobile app to more than 150 customers this fall. The app provides expanded information about a railcar’s whereabouts, such as notifying a customer when there’s a network issue that may delay a shipment. It explains the reason for the delay and outlines available solutions.
The railroad will expand the app to its entire customer base in 2025.
UP is also exploring ways to combine shippers’ technology with the railroad’s resources, enabling AI recommendations and predictions that help assist with customers’ transportation planning. This more closely aligns customers’ supply chains with the railroad’s service.
“For example, AI can be taught to predict shipping patterns that help Union Pacific proactively supply the necessary equipment at appropriate levels to meet our customers’ anticipated needs,” says Tysver. “Another example of how this can be leveraged for the future is in managing weather impacts; specifically, when we know an area is likely to be impacted by weather, we can notify customers and suggest possible re-route options.”
Rob Stevens, vice president at First Analytics, helps railroads implement machine learning, generative AI, and analytical-based software. The tech firm offers services targeting predictive analytics, forecasting, fuel conservation, and any business process tied to software.
Stevens highlights a relationship between AI and preventative locomotive maintenance. Telemetry data from locomotive manufacturers or industry tech companies feeds information to railroads, helping them decipher maintenance trends among certain locomotive models before a line-of-road failure.
“Going through that data,” he says, “and building AI models that are predictive models that say ‘hey, this locomotive may have a stall or a fault, you don’t want that to go down on the mainline and gum up the network’ — I think has become the poster child for AI.”
Safety cannot be understated, either. Stevens says First Analytics got its start in the rail industry by using predictive analytics and machine learning to identify risks linked to human factor injuries or reportable injuries to the Federal Railroad Administration.
“We’re trying to have AI look at the situations that might put this particular crew more at risk of having a human factor derailment,” he says.
Using wide swaths of data, AI identifies trends among derailments and workplace injuries that allows the railroads to get in front of potential problems.
“By employee, for example — what’s their tenure, how much time in the craft, have they had the most recent training, have they had rules violations, drug and alcohol test issues, have they had prior incidents?” says Stevens, highlighting ways data can be sliced using railroads’ dozens of internal databases.
Artificial intelligence is here to stay, and as railroads’ databases become more aligned, railroads are likely to become increasingly reliant on the tech for safety, operations, and planning.
UP, in particular, says AI will continue to be integrated across the railroad.
“We are taking a very surgical, rational approach to deploying AI-driven solutions, making sure it is safely and securely done,” Tysver says. “We anticipate continuously evaluating the potential for AI over the next several years, as we explore and find new ways to improve our business and network fluidity.”
John has been there and seen it. Glad I’m retired from there.
Given that U.P management has little intelligence, its way out is with artificial intelligence.
So glad I retired before this “A1” thing hit.
This is just a use of operational data to improve logistics. Where I think the use of data would have a huge benefit is to perform an analysis of its collective ROW’s.
Have it look for elevation changes, curves or physical impediments that impact fuel consumption, train lengths, car makeups. Align their capital planning to more than just welded rail and replacing ties, but optimizing some of their ROW’s.
I laugh when I see some ROW’s that for more than a mile still follows the small hills and bumps in the geography, which means it hasn’t been changed since the tracks were laid in 1868.
With long trains now the rule, ROW consistency will now become more important. Removing various elevation changes of less than .25% over several miles can have a tremendous impact of fuel consumption and stress at the coupler.
“AI” is not in any way, shape, or form any sort of “artificial intelligence”. What it is is a search engine–it searches throughout either a supplied database or the Internet. It is unlikely to use trains.com articles in its search but instead to use whatever the heck it finds. For all we know it could find some stupid book ways to disrupt the economy. What “AI” then does is it compiles all the information into a mostly coherent statement. It is likely to be inaccurate (for example, it got ‘BNSF 1988’ mixed up with ‘BNSF 198’), and so are its sources. Image generator “AI” can’t generate images of modern EMDs. I have asked it repeatedly to make images of SD60Ms, SD70Ms, SD70ACes, and the like. It consistently produced an often-distorted image of a GE with a metal can instead of a coupler. And this technology is what the UP is using now. I expect more Union Pacific incidents soon; hopefully I’m wrong.
All due respect, what you’re using it for vs. UP’s utilization is apples & oranges. And no one said they’re using AI to make ALL decisions. This is machine learning – just like you wouldn’t entrust dispatching to a human with zero experience, no company is going to turn it all over to an AI algorithm.