The marriage of artificial intelligence and and industrial engineering – or industrial IoT – is promising to disrupt the global economy. If the locomotive industry, for instance, uses smarter machines to offer incremental improvements – even just a 1 percent gain in productivity –  massive economic benefit can be derived for the entire global economy, to the tune of hundreds of billions of dollars.

The stakes are high for the the transportation industry, says Wesley Mukai, chief technology officer at GE Transportation (who will be joining a host of other AI experts at VB Summit: Riding the AI Wave, October 23 and 24 at the Claremont Resort in Berkeley.) For every 1 mph rise in velocity, the industry gains $2.5 billion in value; for every 1 percent improvement in railcar terminal dwell they save $2.2 billion.

And now when a freight locomotive can be loaded with over 200 sensors streaming data to machine-learned analytic applications that process over one billion instructions per second, those numbers are becoming real-world results, as the promise of AI-powered industrial IoT kicks in.

AI on the edge

“I always think about it as edge to cloud,” Mukai says. “Enabling even the potential for AI really starts at the locomotive.”

Each of GE’s smart freight locomotives is equipped with what is essentially a mobile data center on the locomotive plus physical, high-definition cameras that capture track, front and back, and the cab of the locomotive. The data is aggregated right there on the edge gateway and analyzed, enabling near real-time decision-making on the locomotive itself.

For instance, much of the on-board video analytics are dedicated to the same things that automated car sensors are — instead of the road, they’re inspecting the track, identifying signs and mile posts, and scanning for obstacles on the track. But the biggest impact, Mukai says, is the ability to identify sun kinks, or track deformations, that lead to derailment. Video physically captures track flaws; machine-learned models can detect whether they’re a danger to the locomotive.

“The impact has been incredible,” say Mukai. “We’re seeing great gains there in terms of speed, because accuracy detecting things on track or things around the track is up to 99 percent because of AI.”

Major impact in the cloud

But the heavy, scaled computing happens in the cloud, so these new locomotives combine edge capability with cloud connectivity, says Mukai, in order to tackle the advanced analytics.

At GE, the data is analyzed at their Global Performance Optimization Centers (GPOC) where predictive machine-learned algorithms continue to develop the future forecast and build strategies to optimize everything from fuel use patterns to maintenance schedules.

Locomotive health — the factor that an enormous part of the transportation network hinges on — has been revolutionized with sensorized machines. They’re going beyond traditional models that alert operators and maintenance crew to coolant levels and the like. Advanced anomaly detection can be extrapolated to predict and circumvent failures. Which means they can be caught before a train can cause a backup from Chicago to New York and a tsunami of other effects across the entire ecosystem, as track congestion directly impacts throughput and velocity of freight across the system.

Their early pilot with Deutsche Bahn in Germany has seen a 25 percent reduction in locomotive failure rates, which is extraordinary, Mukai says.

Ripple effects

They’re also seeing major gains in the optimization of their trains and transportation network, Mukai says, turning coordinating schedules from a big operations research problem into an algorithm. Arrival times can be predicted close to the minute based on actuals and a range of other inputs including weather and traffic data, correlating a galaxy of data to pump out a better route or schedule.

Even if the receiver at the end of the route knows it’s still going to be two days and seven hours from now, they know exactly when to expect their cargo, meaning that they’re prepared with the right equipment, workforce, trucks, carriers and rare ready to go.

“Improving the estimated time of arrival improves the throughput across multiple modes of transportation,” Mukai says. “By being predictable, we’re reducing so much cost in the system.”

Workforce optimization with automation

For mining operations in remote areas, where the routes are repetitive and the obstacles few, automated trains are already a part of the landscape. But putting automated trains in general production is not far away. To create an automated train that can take on a major transportation hub, you need a variety of technologies in place. Regulation and safety precautions are also essential, but what the industry is moving toward is getting to zero-to-zero functionality.

You already see remote control driving of trains within a yard, streamlining the work previously requiring many people into something a single person can do with what’s essentially a handheld remote control unit. You can actually drive the locomotive and move it backwards and forth at a lower speed in order to manipulate it.

“And then with more traditional freight, you see this sort of stair step progression of technologies applied to balance safety and fuel economy,” Mukai says.

So drone trains aren’t on major trade routes yet, but AI-powered automation is already creating big gains for traditional freight trains operated by the living and breathing. Tools like GE’s trip optimizer can predict environmental conditions, update optimized trip plans, and warn the operator to adjust velocity at any moment.

It can remind a driver, for instance, that instead of pushing his engine hard to the crest of a big hill, he can ease up on the throttle a certain distance before the crest and let momentum carry him over.

“With tricks like this, the optimizer can save operators tens of millions of dollars in diesel,” Mukai explains.

He adds that in the self-driving train space, it’s not always about trying to take away the whole crew — with automation and a single-person crew, a highly trained workforce can be distributed more effectively.

The future of AI-enabled industry IoT

Network coordination is the missing piece, Mukai says. For instance, a consistently updating optimization plan should be able to connect to the trip optimizer and adjust automation as necessary; a movement plan should be able to optimize both the yard and the main line at the same time. All personnel should be able to see where the trains are and which should have precedence, in a system as fully integrated across the line as air traffic control boards are between airports.

“In our innovation accelerator, a lot of the work is around the integration of the applications between many of these traditionally siloed parts of the ecosystem, so we’re developing command center applications,” he explains. “And then looking for new ways to visualize that information and interact with those users.”

Staying competitive in the transportation space

The big leaps in car automation may be a red flag for the rail industry, Mukai says. Automated trucks are closer than ever, and not only are they ahead of self-driving cars by utilizing convoys with the lead vehicle controlled by a human, but they’re threatening the precedence of train transportation with rapid-fire innovation.

That’s because trucks are constantly updating their fleets, enabling them to continuously improve technology — a three-year span before retirement instead of a 30-year span which has been typical in locomotives. The business model in terms of safety and insurance is a no-brainer, he adds, and regulations are actually being designed to push the automated vehicle fleet dream faster toward reality.

“With automatic trucks, just think about the efficiencies and the fuel savings — you could really clear up a lot of the traffic, but also improve the supply chain,” Mukai says. “How can the rail side combat that? They need to adopt technology faster.”