AMD Powers AI Track Inspection Solution for Japanese Bullet Train Operator JR Kyushu

[btarunr]

AMD today announced that JR Kyushu Railway Company (JR Kyushu), a Japanese bullet train operator, is using the AMD Kria K26 System-on-Module (SOM) to automate track inspection. This AI-based solution replaces traditional methods of inspecting miles of track on foot, creating significant efficiencies by improving the inspection speed, cost and accuracy to meet Japan's stringent railway safety requirements. JR Kyushu bullet trains operate across an enormous territory of more than 1,455 miles of railroad tracks, with trains traveling at speeds of up to 161 mph. Safety is the company's number-one priority, requiring diligent track inspections carried out at specified intervals.

To increase the efficiency and accuracy of assessments, JR Kyushu chose an AMD-powered solution from Tokyo Artisan Intelligence (TAI), which uses high-speed image processing and advanced AI capabilities to detect and inspect loose bolts and other track issues. "With the new solution from TAI and AMD, we were able to improve the efficiency of conventional track inspection, and we anticipate further improvements in inspection efficiency through future functional enhancements," said Kazuhiro Sakaguchi, deputy manager in the Engineering Division, Shinkansen Department at JR Kyushu.



At the heart of the track inspection solution is a vision computing box attached to a cart that inspects tracks at speeds of 12 mph. The box features a high-speed camera that uses the FPGA-based Kria K26 SOM for AI-enhanced pre-and-post data and image processing. The Kria K26 SOM is a compact, all-in-one embedded platform that integrates a custom-built AMD Zynq UltraScale+ MPSoC with DDR memory, nonvolatile storage devices, a security module, and an aluminium thermal heat spreader.

"The most important benefit of AI in this case was reduction of cost," said Hiroki Nakahara, co-founder and CEO of TAI. "Replacing the conventional method of inspecting tracks on foot with carts has led to a dramatic improvement in operational efficiency."

The programmability, durability, and embedded intelligence of the AMD Kria SOM-powered solution also make it adaptable to the unique and changing conditions, geographies, and needs of JR Kyushu's territory and ridership. Since railways are installed in a natural environment, the ability to update the Kria SOM to suit the daily changing natural conditions is critical and helps future-proof investments.

"AMD Kria SOMs are accelerating innovation at the edge and simplifying solution development through deployment," said Chetan Khona, senior director of Industrial, Vision, Healthcare and Sciences Markets, AMD. "JR Kyushu is a perfect example of how the endless programmability of Kria SOMs, combined with edge AI computing, can automate processes and dramatically enhance operational efficiency across a variety of applications, from machine vision to industrial robotics and AI/ML computing."

Read the AMD and JR Kyushu customer case study here.

View at TechPowerUp Main Site

 

[Fluffmeister]

Track inspectors in Japan out of work, thanks AMD.

 

[Carillon]

If the speed of the cart is limited by the speed of the NPU, an upgrade to this AI accelerator would literally accelerate the cart!

 

[Wirko]

A camera that can detect loose bolts? I want one of those too.

 

[Ferrum Master]

Track inspectors in Japan out of work, thanks AMD.

Fukushima had another small leak meanwhile, how about giving away life depending infrastructure to machines, as those don't do hangovers or have morning quarrels with their wives, thus doesn't close some valves.

No job is really taken away as somebody also has to do hardware maintenance, it just increases redundancy.

 

[Panther_Seraphin]

What I suspect is that they will be able to reduce the amount of time an area of track is out for maintenance with these things.

Also if you still wanted a human to follow up on all of the track checked by these devices the same as now, it could be used to just get repair crews in quicker to get to areas that definately need work while the rest of the track is inspected with current methods as well.

 

[cvaldes]

What I suspect is that they will be able to reduce the amount of time an area of track is out for maintenance with these things.

Yes, this is the point.

All Shinkansen tracks are examined nightly in the wee hours after midnight once service ends. Using AI assisted track inspection devices will speed up the process and potentially offer longer service hours.

It's this level of safety that has helped the Shinkansen maintain a superb operational record. There have been zero track fatalities since the first line opened in 1965.

It's dark and often cold when these inspections are done, sometimes including rain, sleet, and snow. Not particularly the ideal conditions that most humans would prefer. But that's the way they are done to impact service the least. This is precisely the type of situation where AI assistance can help the humans running the machines. Of course, the track inspection instruments they have been using have been upgraded over the years. This is just another iterative improvement in the process.

Note that technological improvements have also helped increase bullet train speeds over the years without compromising safety or comfort.

 

[Wirko]

In the future they will probably have real-time track inspection systems on the trains themselves, with cameras and other sensors. Why not? At 320 km/h versus 20 km/h, sixteen K26 computers would suffice for the job. Cameras would have to be pretty advanced and very fast but the Japanese know how to make cameras, I guess.

 

[Panther_Seraphin]

In the future they will probably have real-time track inspection systems on the trains themselves, with cameras and other sensors. Why not? At 320 km/h versus 20 km/h, sixteen K26 computers would suffice for the job. Cameras would have to be pretty advanced and very fast but the Japanese know how to make cameras, I guess.

I can imagine these things would be used to priorities areas where things have been flagged for concern.

Passenger train goes across, system notcies an issue thats non safety critical. Moment the track is closed that night for normal maintenance while the "Doctor Yellow" trains start moving you can get human eyes and ears directly to those areas to verify straight away.

 

[cvaldes]

The AI system will also be used when human safety is of particular concern. How would you like to visually inspect an elevated track at 2am during a typhoon? Many of the shinkansen tracks are above grade, especially in urbanized areas.

 

[Panther_Seraphin]

A good example of using "AI" to HELP worker vs replacing workers

 

[L'Eliminateur]

Track inspectors in Japan out of work, thanks AMD.

that's how life and the technical advances works, deal with it. You can't hold back technological advances just because someone will be out of job.

¡Ban those devil electric lights!, street lighters out of work, thanks Edison.
fire stokers out of work, thanks Electro-Motive Diesel!.

This application is one of the very few things that AI really seems to good and useful and not some bolted-on gimmick or "the cloud"

 

[cvaldes]

that's how life and the technical advances works, deal with it. You can't hold back technological advances just because someone will be out of job.

¡Ban those devil electric lights!, street lighters out of work, thanks Edison.
fire stokers out of work, thanks Electro-Motive Diesel!.

This application is one of the very few things that AI really seems to good and useful and not some bolted-on gimmick or "the cloud"

Yes, the sheer irony of someone going online to rail against the relentless march of technology.

I'm glad I can count on TPU to embrace genius visionaries like them.