Precisely.
Overmod Euclid So if there is a motive to automate inspections, I believe it is the motive to eliminate the cost of employing humans to do the work. Damn, you just won't let go of the stupidity, will you? The point of the automatics is to REDUCE both the frequency and the relative criticality of having human inspections to ensure the right mix of safety and maintenance cost. It is not to REPLACE the humans ... even if the Government went along with this silliness and stopped their safety oversight because 'they trusted the technology'. You might as well try saying that remote maintenance routines on locomotives eliminate a need for blue-card inspections. Try that one on the FRA and see what response they give you. It is possible, at least in theory, that some combination of AI and machine vision might automate the tasks human inspectors do, as completely as they could do it. That would still not replace the importance of periodic human inspection or review, and while this might be guided by overseeing the validity of the data gathered by the automatic systems, it wouldn't be limited to accepting it as complete. I note that in the BNSF timeline, it specifies the reduction of human inspections. I do not see anything there that implies either that these could, or would, be extrapolated to zero as the system proves itself. Now, I'm prepared to be wrong about this, but I have seen nothing so far that indicates the point of the advanced technology is to supplant either human oversight or human 'backstopping', nor would I believe that railroads would put full reliance in that idea. As I think I said: if there are individual inspectors who routinely find 'excess' critical defects that both the automatic systems and subsequent human inspection prove 'not there', it may give railroads much greater incentive and proof to go to, say, the DOT OIG and petition to have different inspectors. I'd expect that to reduce any temptation to impose punitive numbers of 'defects' fairly quickly, and reduce the practical careers of any inspectors so motivated no less so.
Euclid So if there is a motive to automate inspections, I believe it is the motive to eliminate the cost of employing humans to do the work.
Damn, you just won't let go of the stupidity, will you?
The point of the automatics is to REDUCE both the frequency and the relative criticality of having human inspections to ensure the right mix of safety and maintenance cost. It is not to REPLACE the humans ... even if the Government went along with this silliness and stopped their safety oversight because 'they trusted the technology'.
You might as well try saying that remote maintenance routines on locomotives eliminate a need for blue-card inspections. Try that one on the FRA and see what response they give you.
It is possible, at least in theory, that some combination of AI and machine vision might automate the tasks human inspectors do, as completely as they could do it. That would still not replace the importance of periodic human inspection or review, and while this might be guided by overseeing the validity of the data gathered by the automatic systems, it wouldn't be limited to accepting it as complete.
I note that in the BNSF timeline, it specifies the reduction of human inspections. I do not see anything there that implies either that these could, or would, be extrapolated to zero as the system proves itself. Now, I'm prepared to be wrong about this, but I have seen nothing so far that indicates the point of the advanced technology is to supplant either human oversight or human 'backstopping', nor would I believe that railroads would put full reliance in that idea.
As I think I said: if there are individual inspectors who routinely find 'excess' critical defects that both the automatic systems and subsequent human inspection prove 'not there', it may give railroads much greater incentive and proof to go to, say, the DOT OIG and petition to have different inspectors. I'd expect that to reduce any temptation to impose punitive numbers of 'defects' fairly quickly, and reduce the practical careers of any inspectors so motivated no less so.
I believe that the automation is to ultimately replace most, but by no means all, human inspectors. It's in line with all the other places they are using more automation.
They have, or had, a pilot program to use hot box detectors at certain locations to do away with the 1000 (and maybe also the 1500 mile) air brake inspections for certain trains. Less carmen to have to employ.
And the railroads have floated the idea of relieving the requirement of the daily locomotive inspections.
They ain't buying automation just for the heck of it.
Jeff
Here is a pretty straight forward explanation of what the difficulties are in terms of error:
https://www.abtasty.com/blog/type-1-and-type-2-errors/
SALfan BaltACD In one's local area - look at the published inspection reports of local eating establishments - I have yet to see one that has stated NO VIOLATIONS FOUND - and that runs the gammut from the swankiest of high roller establishments to the hole in the wall greasy spoon. Governmental inspectors - no matter the industry - don't retain their jobs if they don't find violations. The number of violations found in restaurant inspections are not necessarily a product of the inspector/restaurant relationship, or the attitude of the inspector. I make my living reviewing the performance of contractors on contracts with a government agency (internal auditor), and recently reviewed a food service contract where contractor personnel and management use government kitchens to produce meals. Take it from someone who spent six months on an exhaustive review of kitchen operations, there are about 862,000 ways to screw up food handling, preparation, serving and cleanup of just one meal (a slight exaggeration, but you get the idea). I watched a meal being prepared and served, and part of the cleanup while accompanied by the contract manager (a former Food Service Director with 10 years' experience), the Regional Manager of the contractor (another former Food Service Director with years of experience). Both of those folks pointed out several things I didn't even notice that were done wrong or could have been done better/faster/easier. It was truly an eye-opening experience how tiny little things could be a violation of the food-service rules, or could possibly lead to violations if the production workers don't do everything just so. And this contractor does a good job; they aren't one of those contractors who make you wonder if they have ever even walked by a kitchen before.
BaltACD In one's local area - look at the published inspection reports of local eating establishments - I have yet to see one that has stated NO VIOLATIONS FOUND - and that runs the gammut from the swankiest of high roller establishments to the hole in the wall greasy spoon. Governmental inspectors - no matter the industry - don't retain their jobs if they don't find violations.
In one's local area - look at the published inspection reports of local eating establishments - I have yet to see one that has stated NO VIOLATIONS FOUND - and that runs the gammut from the swankiest of high roller establishments to the hole in the wall greasy spoon. Governmental inspectors - no matter the industry - don't retain their jobs if they don't find violations.
The number of violations found in restaurant inspections are not necessarily a product of the inspector/restaurant relationship, or the attitude of the inspector. I make my living reviewing the performance of contractors on contracts with a government agency (internal auditor), and recently reviewed a food service contract where contractor personnel and management use government kitchens to produce meals. Take it from someone who spent six months on an exhaustive review of kitchen operations, there are about 862,000 ways to screw up food handling, preparation, serving and cleanup of just one meal (a slight exaggeration, but you get the idea). I watched a meal being prepared and served, and part of the cleanup while accompanied by the contract manager (a former Food Service Director with 10 years' experience), the Regional Manager of the contractor (another former Food Service Director with years of experience). Both of those folks pointed out several things I didn't even notice that were done wrong or could have been done better/faster/easier. It was truly an eye-opening experience how tiny little things could be a violation of the food-service rules, or could possibly lead to violations if the production workers don't do everything just so. And this contractor does a good job; they aren't one of those contractors who make you wonder if they have ever even walked by a kitchen before.
This post triggered a memory of a superb Fawlty Towers episode entitled "Basil the Rat" about a visit from the health inspector.
https://www.dailymotion.com/video/x6thosu
Electroliner 1935 And I am not aware of one man locomotives on Amtrak had had any failures to require a second person in the locomotive.
When it comes to issuing inroute Train Messages to Amtrak - there can be issues with only the Engineer on the locomotive. CSX rules prohibit the person at the opeating controls of a MOVING locomotive from copying a Train Message. The work around is to do it at a station stop or get the Conductor onto the engine.
Never too old to have a happy childhood!
As Larry pointed out, defect severity comes in shades of GRAY. Manual detection is subjective and individual inspectors judgement may think something is not urgent and not worth reporting and/or management may say the same thing or that the budget doesn't have the money this period, it can wait until the next period. So time and experience will tell us what happens.
I remember the dire predictions of reducing train crews to two persons. You would expect there to be thousands of accidents freom what was claimed. I am concerned that one man crews may be the step too far but Indiana RR seems to be working out ok. And I am not aware of one man locomotives on Amtrak had had any failures to require a second person in the locomotive.
Electroliner 1935I hope it will not find so many insignificant issues that management ignores some flaw that causes a wreck with fatalities.
This is precisely the convergence that is the bane of automatic-system designers. Since they can't predict THE flaw that causes the wreck, they have to set k high enough to be reasonably sure to catch it, whatever it may be. And, even then, there's no guarantee the 'flaw' may be a complex interaction, or rapidly emergent from a combination of sub-critical things, or result from some act of God since the last formal check.
A further issue, which we've discussed in various contexts, is how much 'trust' a railroad can put into what the automated systems are reporting. It would be nice to leave it at 'trust, but verify' (which is what a good system here would do) but the problem is precisely in knowing what needs the verification, and having the time and means to do it.
There are of course methods of intermediate data analysis and 'warehousing' that help optimize this, but whether or not they get done on the scale, or with the details, that are needed is a question I haven't had a good answer to or for, yet.
PSA tests have false positives but most of us old geezers still get them. But then have the Dr. decide what to do. Should we ignore them?
Our wives have mamograms which also have error rates. Should we ignore them?
It is easy to be suspicious of management that has a history of being against the worker and some (% not known) management think workers don't care about the job, they just want a bigger paycheck. And I suspect that we all know men who take a lot of pride in doing a great job and others who don't. What I hope comes out of this is a safer railroad. Mixing the benefits of automation to see what it can detect and what it can help manual inspections find. I hope it will not find so many insignificant issues that management ignores some flaw that causes a wreck with fatalities.
EuclidSo if there is a motive to automate inspections, I believe it is the motive to eliminate the cost of employing humans to do the work.
Overmod Euclid You make it sound as though I have offered a blueprint for an automatic inspection machine. What I said was that you HAVE to offer that blueprint, or at least cogent explanation of how you intend to make the solution work, before "concluding" that automatic inspection is "the" answer. And none of us is going to believe you did not say that, in black-and-white terms.
Euclid You make it sound as though I have offered a blueprint for an automatic inspection machine.
What I said was that you HAVE to offer that blueprint, or at least cogent explanation of how you intend to make the solution work, before "concluding" that automatic inspection is "the" answer. And none of us is going to believe you did not say that, in black-and-white terms.
I never said I intended to "make a solution work."
Here is what I said:
IF the human inspection process is as dysfunctional as Balt says it is; That is, with all inspectors adding non-existene defects because they don't want to be fired because management believes that all inspectors find too few defects; if that is true, then no human inspection will work.
And if that is the case, the only workable inspection will be by automatic machines. The machines won't fudge the inpection results in order to keep from getting fired.
But backing up a bit, I don't believe Balt's assertion that all inspectors add false defects in order to keep from getting fired. I suspect that most human inspection is done in good faith with full objectivity and confidence. So if there is a motive to automate inspections, I believe it is the motive to eliminate the cost of employing humans to do the work.
Balt said this: The one thing about 'Inspectors' as a profession, no matter if they are public or private - "If they don't find defects they are fired." ie. if in normal means they didn't find defects, they have to create defects (real or on paper) to go on to the next territory.
EuclidYou make it sound as though I have offered a blueprint for an automatic inspection machine.
All I have done is consider Balt’s assertion of a universal truth saying that the human inspection process is flawed because inspectors worry that they will be fired for not finding enough defects.
We've had people discuss the 'good' and 'bad' FRA inspectors, the kind of relationship you can build with them, the pressures they may encounter from time to time from the 'political' side of the agency ... as I recall, a number of threads with more concrete details from RyPN. There may be things you could 'do' if a given inspector produces more than a historical number of 'defects' -- whether justified in the event or not -- or in getting them removed or reassigned through administrative complaint if it appeared, or turned out, that fewer defects actually required the action 'required'. Either of those would be preferable in either the short run or the long run to very expensive implementation and then what you make out to be blind reliance on an automated system ... which, in order to catch 'all' critical defects, may have to be set at least as 'restrictive' as the worst politically-motivated human inspector.
Based on his assertion, I see the solution as being automatic inspection equipment that management cannot fire. That’s all.
You have now added the worst of all qualifiers: a robot from whose 'decisions' there can be no appeal, running software for which the railroad can have no input.
That is not at all the point of the BNSF program, and I'd argue it's not really the point of any machine-vision or other sensing/detection program I've seen. Part of the idea of the automatics is to generate an enormous volume of raw data at high speed, far faster than any Mk 1 eyeball no matter how enhanced, and then extract likely signatures from that. Think of this as a very fast, much more capable version of what a Sperry car does inductively or ultrasonically in finding hidden rail flaws, extended now to a much wider range of potential issues. Keep in mind that it is possible to design sensor-fused machine vision that could detect motion in the rail or track with multiple passes at different dynamic loading; nobody said this is a replacement for linear trackwalking every week plus crews to the site of a reported riding defect or crew-observed anomaly.
I never said how the equipment should be designed.
B&O, just before WWII, came up with a nifty idea: a forward-feed stoker. This would throw fuel back toward the heel of the firebox instead of from a table at the back necessitating a bunch of shovel work. This looked promising enough that B&O installed the equipment on what might have been more than 70 locomotives. We don't know how many, precisely, because the system did not work, perhaps couldn't have worked in a normal railroad environment of that era, and the whole thing was abandoned so quickly and conclusively that you won't find much mention of it. Existing methods of high-speed machine inspection are really not that much more different: you must know enough about what they say they do before you specify them as THE solution. Or spend a lot of money and cry a lot of tears compared with keeping trackmen on at least their essential part of the job, and maintaining rapport with your FRA people.
Of course BNSF and the FRA would first need to see the false positive and false negative rates of both automated and human inspections.
Overmod Part of what is at the root of this discussion -- not the root, but important enough -- is the idea (espoused by Euclid in particular but shared by some others) that the automated inspection effectively catches all necessary defects. (Leave aside for the moment any discussion of whether it catches 'all defects a human inspector would'). This is essentially a religious belief, not a technical one, and before we can reach any 'black and white conclusion' we have to look a bit at the automated testing and its consequences. There is a metric called the crossover error rate, defined as the point where false positives equal false negatives as the system runs. (You may translate this in Euclid's scheme into the number of phony 'critical-response' events that the enforcers invoke fines and penalties for not redressing 'timely' vs. the number of safety-critical problems that might cause accidents or increasingly expensive track (or train) damage. It would be simplistic to think you could set this 'equal' even if the CRR were small in actual size (which good systems try to accomplish). However, it would be equally simplistic to invoke Euclid's criterion on safe use of the emergency brake and ensure that every defect is found even if that means an exordinately large number of technical false detections. I would like to see how Euclid balances the risk of false-rejection of legitimate defects with his concern that inspectors may be 'finding more defects that may or may not be there' out of a heightened concern for "safety".
Part of what is at the root of this discussion -- not the root, but important enough -- is the idea (espoused by Euclid in particular but shared by some others) that the automated inspection effectively catches all necessary defects. (Leave aside for the moment any discussion of whether it catches 'all defects a human inspector would'). This is essentially a religious belief, not a technical one, and before we can reach any 'black and white conclusion' we have to look a bit at the automated testing and its consequences.
There is a metric called the crossover error rate, defined as the point where false positives equal false negatives as the system runs. (You may translate this in Euclid's scheme into the number of phony 'critical-response' events that the enforcers invoke fines and penalties for not redressing 'timely' vs. the number of safety-critical problems that might cause accidents or increasingly expensive track (or train) damage.
It would be simplistic to think you could set this 'equal' even if the CRR were small in actual size (which good systems try to accomplish). However, it would be equally simplistic to invoke Euclid's criterion on safe use of the emergency brake and ensure that every defect is found even if that means an exordinately large number of technical false detections.
I would like to see how Euclid balances the risk of false-rejection of legitimate defects with his concern that inspectors may be 'finding more defects that may or may not be there' out of a heightened concern for "safety".
The errors that a system can make are divided into two types, helpfully termed 'type-1' and 'type-2' by the Pole and the Brit that first worked on the analysis. (The problem is that there's no indication 'which is which' in the definition, and the definition is exactly backward in the security industry from what it is in logic and statistics... so it's a bit like remembering how to convert from Grays to Sieverts.)
For the moment, let's use system security and call type-1 the false rejection: there was no track defect present, but the system flagged one (for attention, and remediation within the FRA time period, etc.). The type-2 is the false acceptance (there was a defect, perhaps a critical one, and the system said there was not). Automated systems will suffer many instances of both as they operate; I'm sure mudchicken has more than a long list of the criteria).
In statistics this is partly addressed by using what is called the 'utility function' (k) which here weights detection of defects more strongly than 'editing out' common artifacts or causes that make the system "think" there are defects. Some part of fine-tuning this is, very likely, part of what BNSF will be testing -- note that manual inspections are still being called for, at reasonably frequent intervals; there appears to me to be no camel's nose scheming to replace all manual inspections with autonomous sensor-fused intelligence -- just a reduction in the time between careful boots-on-the-ground, and a better idea of areas to give more attention when the boots on the ground get there.
Euclid then you act like they are my ideas.
Not your ideas - your conclusions.
Larry 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...
tree68 Euclid If this (what Balt said about inspectors reporting false defects in order to not get fired) happens in the mindset of every inspection, Balt has made the perfect case for replacing human inspectors with automated inspection machines. Once again, you feel it has to be either black or white. You've shown before that in your mind, there can not be any shades of gray. In real life, there are shades of gray.
Euclid If this (what Balt said about inspectors reporting false defects in order to not get fired) happens in the mindset of every inspection, Balt has made the perfect case for replacing human inspectors with automated inspection machines.
Once again, you feel it has to be either black or white. You've shown before that in your mind, there can not be any shades of gray.
In real life, there are shades of gray.
EuclidIf this happens in the mindset of every inspection, Balt has made the perfect case for replacing human inspectors with automated inspection machines.
+1!!!
tree68 Euclid So if all inspections, including railroad inspections, lead to flawed conclusions due to the personal agendas of the inspectors and their employer—what then is the answer? There are established standards for most any inspection. Most inspectors won't ignore flaws if they value their employment (such behavior has a way of coming back to bite you). The other end of the scale is malicious compliance - writing up every little thing. I'm sure there are folks who will go to either extreme, but as you note, that usually involves an agenda. Of course, there are those fields where you want every little thing written up. No question there. However, the conclusion that all inspections are flawed is erroneous. I hesitate to suggest an example, lest someone take it literally, but perhaps a spike that isn't quite all the way in, when all of its brothers are otherwise fine can be ignored for the moment. There is no compromise to safety or operations, so let it slide. I'm pretty sure that those setting the parameters for automated inspections will have such possibilities in mind.
Euclid So if all inspections, including railroad inspections, lead to flawed conclusions due to the personal agendas of the inspectors and their employer—what then is the answer?
So if all inspections, including railroad inspections, lead to flawed conclusions due to the personal agendas of the inspectors and their employer—what then is the answer?
There are established standards for most any inspection. Most inspectors won't ignore flaws if they value their employment (such behavior has a way of coming back to bite you).
The other end of the scale is malicious compliance - writing up every little thing.
I'm sure there are folks who will go to either extreme, but as you note, that usually involves an agenda.
Of course, there are those fields where you want every little thing written up. No question there.
However, the conclusion that all inspections are flawed is erroneous.
I hesitate to suggest an example, lest someone take it literally, but perhaps a spike that isn't quite all the way in, when all of its brothers are otherwise fine can be ignored for the moment. There is no compromise to safety or operations, so let it slide.
I'm pretty sure that those setting the parameters for automated inspections will have such possibilities in mind.
Answer (freely paraphrased): "Ask Balty. He doesn't trust any data."
Meanwhile, in the real world...
It's been fun. But it isn't much fun anymore. Signing off for now.
The opinions expressed here represent my own and not those of my employer, any other railroad, company, or person.t fun any
BaltACDDon't assume.
Why not? My assumption was entirely correct according to your own explanation. I realize you were talking about all types of inspections. The point is that one of those inspection types was railroad track inspection. And I agree that there are bad motives and mistakes made thoughout every bit of all types of inspections. Therefore, the bad motives and mistakes occur in railroad track inspections. I think that was your point.
So if that is the point, let's get rid of the human element of track inspection because it is a trouble spot. Every bit of track inspection can be done by automatic machines, including machines that measure spike withdrawl.
I don't see how anybody can argue that automation is unecessary when it is the only solution to the problem.
Euclid tree68 Euclid If there really is all of this subterfuge and deception going on to “cook” the inspection results based on distrust within the employee-employer relationship, it seems to me that alone would be a good reason to automate the process, and take the human element completely out of it. I think Balt's original comments suggesting that an inspector isn't inspecting unless he finds flaws was more of a general thing, not specifically railroads. Auto repair shops are famous for that type of thing. I was once told my year old truck needed $1,300 worth of work to pass the state inspection. That was at the dealership where I bought it. Another shop passed it "as is," and that was a shop I definitely trust. That's not to say that there's not an adversarial relationship between management and labor, but I don't think it's as bad as all that when it comes to the inspections. I don't think that you can take the entire human element out of the inspections process - as I read it, BNSF wants to cut down the frequency of human inspections for the moment. That's why pilot programs are run - to work out the bugs. If they find that automated inspections are better than human inspections, so be it. You can bet the FRA will be looking over their shoulders very closely. OTOH, they may find that human inspections are actually better. You never know. Balt said this: The one thing about 'Inspectors' as a profession, no matter if they are public or private - "If they don't find defects they are fired." ie. if in normal means they didn't find defects, they have to create defects (real or on paper) to go on to the next territory. While this may apply to lots of different situations and professions, I assume Balt was applying it to railroads. And if it does apply to railroads, it is a dangerous practice that should end. That dangerous practice perfectly justifies ending it by the use of automation. If there are shenanigans committed in the process of inspections, they are committed by humans. So what better solution than to take the human element out of it?
tree68 Euclid If there really is all of this subterfuge and deception going on to “cook” the inspection results based on distrust within the employee-employer relationship, it seems to me that alone would be a good reason to automate the process, and take the human element completely out of it. I think Balt's original comments suggesting that an inspector isn't inspecting unless he finds flaws was more of a general thing, not specifically railroads. Auto repair shops are famous for that type of thing. I was once told my year old truck needed $1,300 worth of work to pass the state inspection. That was at the dealership where I bought it. Another shop passed it "as is," and that was a shop I definitely trust. That's not to say that there's not an adversarial relationship between management and labor, but I don't think it's as bad as all that when it comes to the inspections. I don't think that you can take the entire human element out of the inspections process - as I read it, BNSF wants to cut down the frequency of human inspections for the moment. That's why pilot programs are run - to work out the bugs. If they find that automated inspections are better than human inspections, so be it. You can bet the FRA will be looking over their shoulders very closely. OTOH, they may find that human inspections are actually better. You never know.
Euclid If there really is all of this subterfuge and deception going on to “cook” the inspection results based on distrust within the employee-employer relationship, it seems to me that alone would be a good reason to automate the process, and take the human element completely out of it.
If there really is all of this subterfuge and deception going on to “cook” the inspection results based on distrust within the employee-employer relationship, it seems to me that alone would be a good reason to automate the process, and take the human element completely out of it.
I think Balt's original comments suggesting that an inspector isn't inspecting unless he finds flaws was more of a general thing, not specifically railroads.
Auto repair shops are famous for that type of thing. I was once told my year old truck needed $1,300 worth of work to pass the state inspection. That was at the dealership where I bought it. Another shop passed it "as is," and that was a shop I definitely trust.
That's not to say that there's not an adversarial relationship between management and labor, but I don't think it's as bad as all that when it comes to the inspections.
I don't think that you can take the entire human element out of the inspections process - as I read it, BNSF wants to cut down the frequency of human inspections for the moment. That's why pilot programs are run - to work out the bugs. If they find that automated inspections are better than human inspections, so be it. You can bet the FRA will be looking over their shoulders very closely.
OTOH, they may find that human inspections are actually better. You never know.
While this may apply to lots of different situations and professions, I assume Balt was applying it to railroads. And if it does apply to railroads, it is a dangerous practice that should end. That dangerous practice perfectly justifies ending it by the use of automation. If there are shenanigans committed in the process of inspections, they are committed by humans. So what better solution than to take the human element out of it?
Don't assume. Remember it becomes 'ass U me'. In the 'inspector' profession I was including every form of inspection - building, health, Qualtiy control, process control, weed weasels and it applies to all forms of human endeavor where 'defects' can have consequences far outweighing their individual failure.
If there is a derailment where the rail turned over because poor tie conditions would not hold the spikes - the question becomes why wasn't the condition detected and corrected prior to the derailment.
The continual refrain from senior divisional officials to the subordinate divisional officials - there aren't enough failures on your efficiency tests. We keep having incidents of rule violations that end up in serious incidents, yet your efficiency tests indicate that everyone tested is complying with the rule - the number of serious incidents indicates that there are serious rule compliance failures. Catch them.
In tree's example - the dealers 'inspectors' were more salesman than inspector. In every case of inspection somewhere along the line there is a motive, knowing the motive puts the results of the inspection in its proper perspective.
Our community is FREE to join. To participate you must either login or register for an account.