Artificial Intelligence: A look at how BNSF creates a safer, smarter railroad | BNSF

Highlighting our commitment to safety, service, communities and our heritage.


Artificial Intelligence: A look at how BNSF creates a safer, smarter railroad

Imagine you’re driving and without warning a tire blows. Or, it’s dark and you hit a giant car-crunching pothole. Driving hazards like these are not uncommon, but if you could, wouldn’t you want to know if and when to expect – and avoid – them altogether?

This kind of predictability probably isn’t too far down the proverbial road for drivers. At BNSF, we’re already applying predictive technologies, using artificial intelligence (AI), to find defective track and freight car components, both of which could potentially contribute to train incidents.

While predictive models aren’t new for our 170-year-old railroad, they are now much more accurate, thanks to the speed of today’s computers and Big Data – the massive amounts of information that are now being collected. As a result, AI is transforming many industries, and at BNSF it is one of the top technologies that we are pursuing to not only improve safety, but also asset utilization, service, and operational efficiency.

Just like online retailers and streaming services use AI to make purchasing and TV viewing recommendations, we’re leveraging AI to recognize patterns as well. In our case, AI is helping to better predict issues with equipment components and track so corrective actions can be taken before the failure.

“Simply put, AI is helping us turn our employees from finders into fixers, working to prevent a problem,” said Vice President Technology Services and CIO Muru Murugappan. “These advancements in technology are not only making us a safer railroad, but better able to deliver on our service promise.”

Here are two examples of how AI is making BNSF a safer and smarter railroad.

 

Where there's heat

BNSF has long used sensors – thermal, acoustic, visual and force – positioned along the track to detect freight car wheel defects (locomotives usually aren’t the culprit). Thermal/infrared scanners, for example, can detect the smallest temperature changes of various components. A hot temperature can indicate the brakes are sticking and the components are overheating, potentially leading to a broken wheel.

If the part is already hot and failing, an alarm is sent so the train can be stopped and the car in question inspected. At that point, the car may be monitored until the component can be repaired or the car is removed from the train. While this process takes the bad actor out of the picture, it also means a lot of time is spent making inspections and disconnecting the car, thus delaying the train and impacting our service.

Today, BNSF is leading the industry for using AI for detecting wheel defects as algorithms sort through more than 35 million readings every day taken from our 4,000 wayside sensors to identify potential issues. We use this data to determine the urgency of equipment repairs and to spot trends that indicate when maintenance should happen.

“For example, our data scientists, in cooperation with our mechanical experts, use machine learning to evaluate need-to-watch wheel issues, identifying the conditions that have the highest probability of becoming those likely to cause an incident,” said Matt Baldwin, general director, Cars. “What we find gives us the opportunity to monitor and repair the wheel conditions on those freight cars, likely reducing incidents and service interruptions.”

A newer technology that BNSF is using to identify cracks and breaks in wheels is a machine vision system. Similar to how a child is taught to spot a cat or a dog by being shown pictures, we are teaching AI models to detect broken/cracked wheels by analyzing more than 500,000 images of wheels per day, captured by seven machine vision systems deployed across BNSF’s network. Working with Microsoft, BNSF plans to continue to expand the machine vision program.

“Microsoft is excited to work with BNSF in identifying wheel defects by processing images and data in Azure, our cloud-computing service, thus reducing the time to detection, and ultimately making railroads safer,” said Carlton Dossman, vice president, Microsoft.

 

Looking deep inside

When it comes to our track integrity, we rely on technologies like ultrasound, radar and machine vision systems that look deep inside rail and supporting crossties to find tiny flaws imperceptible to the human eye. These technologies are on manned and unmanned track geometry rail cars that travel across our system to register track wear and tear. Like the freight car with issues, when track has a defect, it means taking that stretch of the railroad out of service or slowing trains down until the track can be repaired, again impacting operations.

The geo cars send track data back to our technology team, who then leverages AI to effectively analyze the hundreds of millions of bytes of information that help drive our track maintenance. A tagging system, for example, is used to identify track conditions. A yellow tag is a “warning.” Now, with the AI data, we can predict which of the yellow tags have the highest probability of turning into red tags – needing immediate attention – within the next 30 days. Today, we classify them as orange tags, elevating the need for maintenance before the track is taken out of service.

Overall, these advanced predictive technologies provide our Engineering teams with a comprehensive maintenance plan. “In the past, most of our maintenance was more reactive to defects,” said Frank Moffitt, general director, Maintenance Planning. “Now, with artificial intelligence modeling technology, we can more precisely know in advance where our efforts are needed most and proactively plan.”

Most importantly, as we get smarter, our network gets safer. And that’s something everyone on the road – rail or otherwise – can appreciate.