Trains.com

Norfolk Southern installs "new" detection devices near the recent East Palestine derailment

6722 views
49 replies
1 rating 2 rating 3 rating 4 rating 5 rating
  • Member since
    January 2014
  • 8,221 posts
Posted by Euclid on Thursday, November 2, 2023 5:58 PM

tree68

 

 
Euclid
QUESTION:  When the system finds a problem, does it make the decision to stop the train?  If so, does the system stop the train?  If that is the case, do the humans decide whether the system command to stop the train is correct?  If the humans conclude that the stop command is incorrect, do they then intervene and overrule the system?   Or--- does the system merely send its findings to the humans to be used as their diagnostic data, with decision of whether or not to stop the train being left entirely up to the humans in all cases.

 

I think the question is - how often are defects detected that require stopping the train?  Is the cost of the detection/reaction system worth the payback?

The problem at East Palestine was not a detector failure, or even a human failure, as such.  It was a failure of the parameters used to determine the severity of the problem.

 

I suspect that the majority of defects found do not require stopping the train.  But clearly the fanfare associated with this new AI based detection is about a comprehensive system that is focused on ultimately preventing derailments. 
 
It is all about safety, and so it focuses mostly on the greatest dangers.  The whole premise of this new technology is that the current protection is lacking.  Certainly if there is a bearing failure that is progressing to the point of derailment, the train should be stopped before it reaches that point. 
 
There should be no sense of trying to race time by attempting to limp into an ideal location in order to get into the clear and not delay traffic.  One point of this new AI system is that if defects are found sooner, there will be more time to get them into the clear without risking a derailment.
 
I don’t know what happened with the East Palestine detectors.  But the company is likely looking at East Palestine as a symbol of insufficient safety, and they are offering this new system as a symbol of the safety solution, ultimately aimed at preventing derailments. 
 
But to East Palestine alone, I read many explanations of the detector problem.  If the problem was a failure of the parameters used to determine the severity of the problem, what was the reason for that failure? I wonder if the parameters were set too tolerant in order to avoid stopping the trains. 
 
Actually, I am not convinced that the East Palestine derailment was caused by a failure of an overheated bearing.  As I recall, the NTSB did initially conclude that the hot bearing was the cause of the derailment.  But they sounded rather tentative.  A hot bearing can cause a derailment, and the train had a hot bearing.  But that does not close the case.
  • Member since
    December 2001
  • From: Northern New York
  • 25,020 posts
Posted by tree68 on Thursday, November 2, 2023 1:20 PM

BaltACD
Remember - PTC is not installed on ALL lines

I thought of that.  Didn't mention it, but it is a complication to the scheme.

LarryWhistling
Resident Microferroequinologist (at least at my house) 
Everyone goes home; Safety begins with you
My Opinion. Standard Disclaimers Apply. No Expiration Date
Come ride the rails with me!
There's one thing about humility - the moment you think you've got it, you've lost it...

  • Member since
    May 2003
  • From: US
  • 25,292 posts
Posted by BaltACD on Thursday, November 2, 2023 9:17 AM

tree68
 
Euclid
QUESTION:  When the system finds a problem, does it make the decision to stop the train?  If so, does the system stop the train?  If that is the case, do the humans decide whether the system command to stop the train is correct?  If the humans conclude that the stop command is incorrect, do they then intervene and overrule the system?   Or--- does the system merely send its findings to the humans to be used as their diagnostic data, with decision of whether or not to stop the train being left entirely up to the humans in all cases. 

I think the question is - how often are defects detected that require stopping the train?  Is the cost of the detection/reaction system worth the payback?

The problem at East Palestine was not a detector failure, or even a human failure, as such.  It was a failure of the parameters used to determine the severity of the problem.

When hotbox detectors first came on the scene, they were usually equipped with a warning light which the crew in the caboose could see.  Then came radios, announcing the train's status.  Even now, I believe a defect detector will announce an issue when it detects it, not waiting until the end of the train (as occurs if there is no defect detected).  This is received directly by the head-end crew.  It doesn't have to go through any back offices, the dispatcher, etc.  There are rules in place for handling such cases.  

So the problem, as you suggest, is finding the happy medium between stopping the train for certain defects (which occurs now, with the human interface), or proactively stopping the train automatically.  Either could cause havoc on a busy line.

Let's take flat spots, for example.  There are established rules for how to handle a flat spot depending on its size, ranging from no special actions to stopping the train in place until the wheelset can be replaced.  Can a lineside detector detect the difference?  Does a detector know the difference between a loaded car and an empty car, which will have different signatures for the same sized flat spot?  Or does that have to go through a back office for decision, human or automated?

With PTC, I suppose the capability exists to enforce such a stop, assuming good communications all along the data path.

All of this puts us one step closer to unmanned trains.

Remember - PTC is not installed on ALL lines

On CSX when I was working - HBD and Dragging equipment detectors report the results of their inspections directly over the Road Radio Channel for the territory the detector is located on.  

WILD (Wheel Impact Load Detectors) are nowhere near a numerous on the railroads as are the HBD/DE installations.  The WILD are monitored by the Mechanical Desk in Jacksonville who issue Mandatory Directives through the CADS to both the Trick and Chief Dispatchers desks where the train is operating.  Trick Dispatcher notifies the train to stop and inspect the offending car(s).  Levels 3 and 4 are the defects are stoppable.  Level 3 upon inspection is allowed to move to destination at 30 MPH.  Level 4 must be set off at the nearest set out track and not exceed 10 MPH getting there.

When HBD/DE are activated, the crew gets the radio announcement and stops train for inspection.  If a Hot Box is found, it is the Conductors decision if the offending car(s) can be moved to a set off location.  At the time I was working the S word was paramount and in many case the Conductor would report the car(s) were not safe to move and the Car Department would be dispatched to change out the wheel set(s).

Never too old to have a happy childhood!

              

  • Member since
    December 2001
  • From: Northern New York
  • 25,020 posts
Posted by tree68 on Thursday, November 2, 2023 8:04 AM

Euclid
QUESTION:  When the system finds a problem, does it make the decision to stop the train?  If so, does the system stop the train?  If that is the case, do the humans decide whether the system command to stop the train is correct?  If the humans conclude that the stop command is incorrect, do they then intervene and overrule the system?   Or--- does the system merely send its findings to the humans to be used as their diagnostic data, with decision of whether or not to stop the train being left entirely up to the humans in all cases.

I think the question is - how often are defects detected that require stopping the train?  Is the cost of the detection/reaction system worth the payback?

The problem at East Palestine was not a detector failure, or even a human failure, as such.  It was a failure of the parameters used to determine the severity of the problem.

When hotbox detectors first came on the scene, they were usually equipped with a warning light which the crew in the caboose could see.  Then came radios, announcing the train's status.  Even now, I believe a defect detector will announce an issue when it detects it, not waiting until the end of the train (as occurs if there is no defect detected).  This is received directly by the head-end crew.  It doesn't have to go through any back offices, the dispatcher, etc.  There are rules in place for handling such cases.  

So the problem, as you suggest, is finding the happy medium between stopping the train for certain defects (which occurs now, with the human interface), or proactively stopping the train automatically.  Either could cause havoc on a busy line.

Let's take flat spots, for example.  There are established rules for how to handle a flat spot depending on its size, ranging from no special actions to stopping the train in place until the wheelset can be replaced.  Can a lineside detector detect the difference?  Does a detector know the difference between a loaded car and an empty car, which will have different signatures for the same sized flat spot?  Or does that have to go through a back office for decision, human or automated?

With PTC, I suppose the capability exists to enforce such a stop, assuming good communications all along the data path.

All of this puts us one step closer to unmanned trains.

LarryWhistling
Resident Microferroequinologist (at least at my house) 
Everyone goes home; Safety begins with you
My Opinion. Standard Disclaimers Apply. No Expiration Date
Come ride the rails with me!
There's one thing about humility - the moment you think you've got it, you've lost it...

  • Member since
    January 2014
  • 8,221 posts
Posted by Euclid on Thursday, November 2, 2023 6:54 AM
Overall, to the extent that the detection system is perfected, the need for human verification of the warnings becomes less necessary.  At some point in this reduction of necessity, the benefit of human verification reverses, and poses more risk of human error than is justified by human verification.
 
I have no doubt that the new equipment is much better than what is in use now.  One might also conclude that if you add human oversight to the automatic system, it can only improve the already excellent latest generation of detection equipment.  I don’t believe that is necessarily true unless it includes the condition that the human oversite cannot overrule the decision of the automatic system. 
 
However, if the humans are allowed to overrule a stop decision by the automatic system, based on the human assumption that the system is incorrect; humans will inevitably occasionally be incorrect and thereby cause a derailment that would not have happened if the decision of the automatic system were allowed stand.   
 
The underlying problem is that there are conflicting objectives in the detection process.  One objective is to prevent derailments, and the other is to not delay trains.  Generally, I believe management worries more about the latter than the former.  So if you put humans into this decision making process, they are going to be seen as being charged with making sure the automatic system warning is correct.  I believe this responsibility will bias the humans to ere on the side of not delaying trains.  
  
QUESTION:  When the system finds a problem, does it make the decision to stop the train?  If so, does the system stop the train?  If that is the case, do the humans decide whether the system command to stop the train is correct?  If the humans conclude that the stop command is incorrect, do they then intervene and overrule the system?
 
Or--- does the system merely send its findings to the humans to be used as their diagnostic data, with decision of whether or not to stop the train being left entirely up to the humans in all cases.
 
If humans are supposed to review data and decide whether to stop the train, will they be immune from management criticism for having caused a false alarm that resulted in stopping the train?
  • Member since
    January 2019
  • 1,686 posts
Posted by Erik_Mag on Sunday, October 29, 2023 2:45 PM

I've most often seen acoustic emissions sensors. Whatever they're called, these coupled with vibration sensors are apparently the "go to" sensors for monitoring bearings. These are typically attached to the external bearing housing for reasons OM mentioned.

The emissions of concern are ~100kHz to 300kHz and come from the dislocations in the metal moving undergoing plastic deformation (yielding).

  • Member since
    September 2003
  • 21,669 posts
Posted by Overmod on Sunday, October 29, 2023 12:08 PM

Acoustic signature detectors are essentially the only technology that can detect 'inside row' roller-bearing degradation prior to catastrophic failure.  A number of the likely failure modes make sound that does not couple well to ambient air, so the wayside detectors need some onboard 'support'.

Checking the inside grease seal for damage -- including from heat and running wear -- is critically important, at every inspection.  While this is impractical to do at wayside runthroughs, the same "AI" technology NS and Georgia Tech have designed could easily be adapted to work with a 'cameras-on-a-stick' gauge slipped in and withdrawn at any rate (software correcting as in those early handheld scanners) during inspection on the ground, or walking the train after a UDE.

  • Member since
    March 2003
  • From: Central Iowa
  • 6,900 posts
Posted by jeffhergert on Saturday, October 28, 2023 5:39 PM

On another site, someone with some experience, says the acoustic detectors are better for detecting roller bearings that are beginning to fail.  A lot better than temperature detectors because by the time some bearings set off a detector, it's almost too late.

He also said working as a car inspector years ago, they would check for a loose seal at the back of the bearing.  They devised/modified a tool to check the seal because it was hard to do a visual inspection.  If the seal was loose, the car was immediately bad ordered and the wheel set changed out.  If not changed out, the loose seal would let the grease leak out and then the bearing would fail. 

Maybe instead of more temperature detectors every 10 miles, there should be an acoustical detector every 50 miles.  And maybe more car inspectors, at least at the major terminals to do a more thorough inspection.

(Instead, as some carmen have intimated, "suggesting" that only one car per shift be bad-ordered, and provided a list of cars that are showing signs of trouble.)

Jeff

  • Member since
    May 2003
  • From: US
  • 25,292 posts
Posted by BaltACD on Saturday, October 28, 2023 3:32 PM

Article states that NS's Portal is being installed a Leetonia, OH which is several miles West of East Palestine.  NS's major Western Pennsylvania yard is Conway which about 20 miles from East Palestine as the crow flies.

My view of the network of Inspection Portals CSX has set up around Waycross  would leave me to believe tha Waycross will become a major car repair location in addition to it being one of the few remaining CSX hump yards.  I suspect this may be the start of NS executing a similar strategy around Conway.

Never too old to have a happy childhood!

              

  • Member since
    September 2003
  • 21,669 posts
Posted by Overmod on Saturday, October 28, 2023 12:52 PM

The AI/ES will compress the information coming off the high-resolution image stream for patterns that match known or anticipated problems.  It will then confirm the severity of the problem, compare any multiple issues, and determine the most effective and safe response, whether that be to put the train in emergency, or use heavy dynamic, or even let the train decelerate to a particular speed.  THAT information will then be passed, with the images and markup, to a human or humans on the 'response team' who would issue the orders to the crew (or other personnel) accordingly.

I would also expect AI/ES to be running on the 'notification chain' for first response, police enforcement, or other contractors, as it can have 'up-to-the-minute' advice without human foreground attention, and handle the necessary communications in order without someone's approval for each communication, with little distraction.

  • Member since
    December 2001
  • From: Northern New York
  • 25,020 posts
Posted by tree68 on Saturday, October 28, 2023 12:35 PM

Euclid
According to this from Progressive Rail, it sounds like humans will decide whether or not to stop the train:

I'd opine that the improved sensor system will provide a better suite of alerts, probably with more strident bells and whistles.  

Just as defect detectors do now, I'm sure problems will be announced in real time, plus the information that will go to appropriate offices at the railroad.  I"m not so sure a direct action without a human interface would be desirable.

LarryWhistling
Resident Microferroequinologist (at least at my house) 
Everyone goes home; Safety begins with you
My Opinion. Standard Disclaimers Apply. No Expiration Date
Come ride the rails with me!
There's one thing about humility - the moment you think you've got it, you've lost it...

  • Member since
    January 2014
  • 8,221 posts
Posted by Euclid on Saturday, October 28, 2023 8:01 AM
Will this new system find a problem, analyze the problem, and make the decision whether to stop the train for further human inspection?
 
Or---will the new system simply gather data in the form of images, temperatures, sounds, etc., and make that data available to human eyes/ears which will evaluate the data and make the decision whether to stop the train for further human inspection?
 
Who or what will decide to stop a train that is indicating a hot bearing?
 
According to this from Progressive Rail, it sounds like humans will decide whether or not to stop the train:
 
The AI transmits the information to the NS Network Operations Center where the data is reviewed by subject-matter experts to identify and address issues. Critical defects found are flagged for immediate handling.
  • Member since
    May 2003
  • From: US
  • 25,292 posts
Posted by BaltACD on Friday, October 27, 2023 4:53 PM

AuTiger19

All three have Waycross in the middle.  Inbounds to Waycross from Savannah, Jacksonville and Atlanta are the prime inspection candidates.

Never too old to have a happy childhood!

              

  • Member since
    July 2010
  • 351 posts
Posted by ns145 on Friday, October 27, 2023 4:41 PM

All this fancy tech is well and good, but the bigger and more complex the installations get, the fewer that will get deployed.  What is really needed is a large number of low to medium cost devices to supplement the data collected by the wayside detectors and provide real time analysis and review capability. 

If NS would have had just one cheap trailcam 10-20 miles west of East Palestine to give the Wayside Desk "eyes" and context on that developing bad journal they could have saved themselves $966 million in direct costs (as of the end of 3Q 2023) and billions in lost Market Cap value.  NS' stock has plummeted from over $255 a share in January to $184.53 today.  I listened to their 3rd quarter earnings call and there's no end in sight on the cleanup costs.  And they are only just starting to receive insurance payments.

  • Member since
    December 2001
  • From: Northern New York
  • 25,020 posts
Posted by tree68 on Friday, October 27, 2023 4:31 PM

The politicians will want one every 10 miles.  Or less.

LarryWhistling
Resident Microferroequinologist (at least at my house) 
Everyone goes home; Safety begins with you
My Opinion. Standard Disclaimers Apply. No Expiration Date
Come ride the rails with me!
There's one thing about humility - the moment you think you've got it, you've lost it...

  • Member since
    December 2005
  • From: Auburn
  • 22 posts
Posted by AuTiger19 on Friday, October 27, 2023 3:25 PM
  • Member since
    May 2003
  • From: US
  • 25,292 posts
Posted by BaltACD on Friday, October 27, 2023 12:52 PM

CSX has had their 'Train Inspection Portal' between Folkston and Waycross operational for the past two years.  I don't know if they are replicating the installation at other locations.

Never too old to have a happy childhood!

              

  • Member since
    September 2003
  • 21,669 posts
Posted by Overmod on Friday, October 27, 2023 11:57 AM

Part of this appears to be an automated end run around the pending Federal agitation for more carman inspections of rolling stock.

24-megapixel cameras take something like 1000 sequential images per car, with the equivalent of broad-spectrum Klieg lighting to make sure the actual resolution and image quality are as high as expected.  These images can be synthetically fused to resolve detail or viewing angle, and I think could reasonably be expected to find most of the potential issues that a carman's visual walkaround inspection might -- with far better documentation and archiving that the work was done.  Another major point (perhaps also reflecting the Georgia Tech IP) is what we were calling for in the wake of the East Palestine accident: AI/ES routines that scan the images for recognizable patterns, then progressions that might be leading to car impairment or accident.  And then notify intelligently, rather than doing the equivalent of finding wolves everywhere in sight.

Of course this does nothing at all about the likely proximate cause of the East Palestine wreck, although the sensor fusion is a potential early step toward the kind of data and metadata transmission architecture that would actually detect progressive bearing failures.

The absolute 'number' of other sensors being rolled out is probably tied to the distance between 'added' facilities -- remember that spacing less than 34 miles apart is now going to be highly desirable if not mandated, with the cobbled-up Congressional expediency congeries currently calling for 10-mile spacing, and perhaps the more "logical" interval being every block of an ABS signal system.

The "near East Palestine" part of the headline probably refers to the first of the 'new' camera arrangements being installed in Leetonia, OH.  Perhaps someone in that area can see if this can be approached or photographed in advance of a more detailed article by NS and Georgia Tech.

  • Member since
    July 2008
  • 2,325 posts
Posted by rdamon on Friday, October 27, 2023 10:43 AM

Prior headilines say 200 additional temperature sensors and acoustic detectors

  • Member since
    December 2006
  • 1,754 posts
Norfolk Southern installs "new" detection devices near the recent East Palestine derailment
Posted by diningcar on Friday, October 27, 2023 10:02 AM

Today's Wall Street Journal has a brief story about more sophisticated detection hardware installed by Norfolk Southern. Does anyone at this site have details.

 

Join our Community!

Our community is FREE to join. To participate you must either login or register for an account.

Search the Community

Newsletter Sign-Up

By signing up you may also receive occasional reader surveys and special offers from Trains magazine.Please view our privacy policy