Hacker News Reader

Comment by njarboe | Hacker News Reader

This article has a lot of text before getting to describing the problem.

Summary: Train wreck 1995 and a installation of a safety-first, CYA response that slows down the trains and does not increase safety.

Longer Summary: Train wreck in 1995. Maybe driver was asleep. Start installing new signals that not only automatically stop the trains when there is a train ahead (good idea and already in place), but also when going over 45mph. Trains used to go up to 55mph with no systemic problems. Drivers go even slower than 45mph because the train is automatically stopped when going faster (as you would want to do if there was a train ahead) instead of just giving a warning and letting the train keep going.

As more of the system gets these new signals, more of the system has slower trains. Thus the increase of delays that are labeled generically as "insufficient capacity, excess dwell, unknown" from about 20% of delays in 2012 to about 60% in 2017.

View on HN

>Trains used to go up to 55mph with no systemic problems.

During the age of steam, express passenger trains regularly exceeded 55 mph. For example, during the 1930s the Burlington Zephyr regularly exceeded 85 mph. For example it made the Chicago to Denver run in 1934 in 13 hours rather than the 18 hrs it takes today.

Source: http://www.slate.com/articles/life/transport/2009/05/stop_th...

This makes me sad about missed potential. I've taken the ski train from Central London to the French Alps - was great. With high speed rail it should be possible to get a 8hr overnight train from Chicago (or Dallas & Houston) to Winter Park CO. I'm guessing it'll never happen

https://www.eurostar.com/uk-en/train/france/ski-train

Running the ski train in Europe makes sense because it's leveraging high speed rail that was built and is mostly used for other things; very little new stuff was needed.

Chicago-Denver, as a city pair, is so sparsely populated and so far apart that new HSR doesn't pencil out, nor does building it out incrementally (it's hard to justify KC to Denver or Omaha to Denver)

We built a rail system that spanned the country once and the highway system doesn't just maintain itself.

That rail system still exists. Asking for a national high-speed rail system is like giving everyone a Rolls Royce.

Imagine if the response to the invention of rail were "well we just built all those canals."

The original rail network was on par with airplanes in terms of relative improvement of speed of travel. The highway network did the same for travel to areas unlikely to ever get a rail connection. (There aren't that many sparsely populated countries with a rail connection to every town; even fairly dense countries don't manage this.)

In today's world, HSR fills a very specific market niche; planes are much faster for long distances due to higher speed and small requirements for land acquisition, and cars are just as competitive for short distances due to the last-mile problem. So I don't think it's reasonable to connect every possible city pair in the US with HSR, but rather it's much better to just build strong regional networks of HSR that work on their own. There's no reason to build a continent-spanning HSR network.

We un-deregaulte rail and force them to offer the passenger service they used to do an excellent job at delivering.

Realistically speaking, you are talking about two different use cases here.

A long distance train benefits from very high top speeds, because it spends a lot of time at those top speeds. Conversely, increasing the high speed of a subway train is subject to the law of diminishing returns, since the train has to comfortably accelerate and decelerate within the span of half a mile. With subways, the main concern is that every single train operates in pretty much the exact same manner WRT acceleration, deceleration, and speed, since the headways on subways can be measured in minutes, or even seconds.

The "super-express" on the A-train is 5.3km and takes 8 minutes, that's 40km/h average (e.g. half of 55mph, with no intermediate stops).

The parent is talking about subway "trains" (sets of coaches, with each car having its own small electric engines, all powered by a common third rail); not, er, train trains (vehicles that pull themselves along two rails using a locomotive.)

Nitpick: self-propelling cars are not a distinguishing property for light rail. Some modern high speed trains use self-propelling cars[0].

[0] https://en.wikipedia.org/wiki/Siemens_Velaro

The technical term is "electric multiple unit", and it's becoming the default for high-speed trains use them:

https://en.wikipedia.org/wiki/Electric_multiple_unit#Example...

https://en.wikipedia.org/wiki/British_Rail_Class_395

High speed aside, in the UK, i think pretty much all inner and outer suburban main-line trains are electric or diesel multiple units (eg if you get a train from London to Cambridge that takes 45 minutes, that's an EMU). According to some random report i found on wikipedia [1], in 2011 there were 1248 locomotive-hauled carriages, 2892 DMU carriages, and 8046 EMU carriages. I could tell you a great many more particulars but suppose that you are tired of it by this time.

[1] http://archive.nr.co.uk/browse%20documents/rus%20documents/r...

Loco hauled are very rare - HSTs from the 70s mainly being replaced on GWR mainline, a little use on WCML and cross country 2 or 3 sets?) and I think the east coast does them too. And Eurostar of course.

Aside from those it's freight and the sleepers isn't it?

If you're counting HSTs then everything on the east coast, plus great Eastern main line, the sleepers, East Midlands HSTs, Chilterns trains, 4 xc sets, the new transpennine sets that are being built, a few Cumbrian coast services, and within a few weeks the great western pocket rocket HSTs and the scotrail ones. Less popular than they used to be, but very rare is overstating it a bit

> Drivers go even slower than 45mph because the train is automatically stopped when going faster (as you would want to do if there was a train ahead) instead of just giving a warning and letting the train keep going.

Is there any protection system that allows the maximum speed to be exceeded with just a warning? Certainly all those I can think of (TPWS, TVM, PZB, or something more modern like ETCS) have a given maximum speed you're allowed to exceed by a small tolerance but nothing more (though, of course, the system may be isolated, but that's not allowed in passenger service).

The failing here sounds like it is setting the line speed to 45mph rather than anything higher.

That, and considering a speed limiter instead of stopping the train when traveling too fast. What good does it do to stop the train when going too fast instead of limiting it in the first place?

The general principle in railway movement authority is that there's an easy fail safe: stop moving. If nothing moves, there can be no collision.

If something goes wrong? Stop the train. You can then figure out what went wrong and address it with no risk. Things shouldn't be going wrong, speeds shouldn't be being exceeded, etc.

Even then you should have some give aka Speed of X+3 MPH break, speed of x+1 MPH warning. You want fault tolerance not simply a hard limit without feedback.

Just a PID speed controller, maybe. Then, some boundary condition where when the controllers inputs mismatched from predicted outputs, then throw an error. But, if you're trying to drive 45, and you achieve 46 with a tail wind it doesn't seem necessary or safer to stop the train.

>If nothing moves, there can be no collision.

If no train on the line moves, sure. But stoping just the one train won't improve safety, and was historically the cause of many accidents. I remember just the one in Shanghai in 201x. One train stopped, the following train hit it.

If stopping doesn't fix things, you're not using enough of it. Time for networked emergency brakes on all trains!

Europe may has that, with the GSM-R system. I remember reading of at least one minor accident that stayed minor because the driver to discover the landslide over the track at low speed passed the button, stopping the incoming faster train on the other track.

http://www.szdc.cz/en/pro-media/tiskove-zpravy/funkce-gsm-r-...

AFAIK, this is just a Czech extension to GSM-R. That said, I recall various instances of GSM-R being used in the UK to broadcast a general message to stop, albeit one that the driver has to act upon.

That shouldn't happen under normal circumstances. That's whole idea behind railway signals.

Rail segments (blocks) are supposed to be automatically locked off while a train is on that segment of track. No two trains can be on the same block at the same time. If a train stops, the signal guarding that block will remain red, ordering the next train to stop before it gets anywhere close.

In fact, the crash at the origin of this whol debacle was caused by a train being stopped on the tracks and being rear-ended by another.

That's an interesting way of looking at the cause and effect. Trains unexpectedly stopping in an expected and inevitable property of all networks, whereas trains exceeding their movement authority is not.

Stopping a train anywhere should therefore always be safe, so the safety failing is with the train behind, not the one in front.

I think the idea is that if the operator is incapacitated or the control system is malfunctioning, stopping the train is the safest solution. While true, a multi-shot timer would be equally effective in detecting such a situation.

There are... one of them is given in the article—which calls the current system "one-shot"—there exists k-shot variants which allow for at least k opportunities/warnings before automagically braking. The (apparent) reasoning behind the MTA not putting these (for some number k>1) was "safety," but the article states that, in general, the timing circuits are much more complicated therefore resulting in a higher cost.[0]

Additionally, a simple case would be to have circuits which do, after a certain point in time, force slow (but safe) braking into the required velocity---I could see problems with robustness here due to the specific application, but at least in theory this should be possible (we all know the difference between theory and practice, though).

On the general case, this all seems a little weird to me since it seems like such a trivial problem (increase error bands, etc), but may have some deeper reasons that aren't explained in the article(?). It's quite difficult to tell since the primary sources are not provided at any point in time.

---

[0] I'm reluctant to accept this claim since most of these things aren't done in an analog way, anymore. Timing is a matter of figuring out empirical constants and programming them in directly, which, while not totally trivial, seems to have roughly the same difficulty (amortized over all of the lights that need to be put in) as doing the timing for a 'one-shot' light.

Indeed, this is my understanding of how most non-US high-speed-rail controllers work - the driver sets a target speed, the physical location on the track has a maximum speed, and the traffic system looks at the cars ahead and sets its own maximum speed (e.g. zero when another train is right in front of you). Instead of stopping the train when the driver exceeds the speed limit, it just uses the minimum of all the target speeds.

TVM is the system I'm most familiar with when it comes to HSR (used on the LGV in France and HS1 in the UK); it has a variable overspeed allowance (5 to 15 km/h depending on current maximum speed). If you exceed that, an emergency brake application occurs.

Japanese ATC and ATS-P, used on all Shinkansen lines and most busy metro/commuter lines, just apply service break until the train is under back speed limit. AFAIK no overspeed allowance.

I think the reason that it's a single shot instead of multi shot and full emergency braking instead of moderate is that the stops are all track based not train based, at least for the 'train in circuit' stops. The original signals basically put up a metal arm that trips the emergency brakes on the train itself so if they piggy backed off this existing system by just adding a speed sensor there's no way to do a slow deceleration because the signals are designed to work without any real cooperation from the train beyond the emergency brake functioning.

Wow. In the Netherlands, intercity trains can go up to 135 km/h (83 mph), and they’ve successfully increased the frequency of these trains on some lines this year. Suffice to say with Dutch population density, the train network is also quite dense. Quite an achievement I think.

The MTA subway isn't really intercity, unless you think of the boroughs as different cities. Although the express trains should be able to go much faster than they do, like the A from 59 St to 125 St, which takes ~8 minutes for 5.3 km even though there are no stops and no other trains on the track.

This kind of thing makes a huge impact on the quality of life for people living in the city.

For intercity trains in the Netherlands some stops are only kilometers apart as well. 50 kilometers between stops would be the absolute maximum.

Yeah. The larger point is still that you just can’t compare IC trains to subways. The TGV gets to 300km/h+, but it’s irrelevant to the Disneyland Monorail.

Many of the subway stops in Manhattan are only hundreds of meters apart, so it is still an order of magnitude of difference.

A closer parallel might be the London Underground, that manages 55-60mph.

The new Elizabeth line is apparently going to go up to 90 miles per hour, but that is probably only for when the train leaves London and heads out to neighbouring towns (it's essentially a subway line that also connects commuter towns to London).

I agree with most of your summary, but I don't believe the article provided evidence that installation of signals to enforce a lower max speed "does not increase safety".

I'd speculate, without evidence, that it probably has increased safety. But maybe it's just a bad tradeoff, compared to other options.

This is one of those comments that is annoying to respond to because it may be technically correct because the original comment wasn't worded ever so precisely, but we all know that "does not increase safety" is not an absolute statement. You could strap a pillow to your chest and get ready to jump out of an airplane, and I'll say "that won't cushion your fall," but "technically, it will," you argue.

Besides--reduced capacity causes platform crowding which is dangerous and displaces passengers into other modes of transportation like foot or automotive traffic, both of which are more deadly per passenger-mile than the subway.

Enforcing the speed limit the system is designed for intuitively sounds like very reasonable and effective safety measure, not comparable in any way to your fantasy scenario.

The whole point of having an analogy in the first place is to sacrifice accuracy in order to illustrate a point in a more obvious way. Analogies are often fantasy scenarios. And making arguments about how things “intuitively sound” is a big step into a world of unscientific thinking.

The question here is, “What is the relationship between train speed and safety?” My guess is that the risk has a bathtub curve to it. At low speeds, certain factors dominate, the dangerous effects of reduced capacity. At high speed, other factors dominate, like crashes. This type of curve is found in many different types of systems. This curve exists in automotive traffic, where both high-speed and low-speed drivers are known to be dangerous, and speed limits that are either too high or too low make the road more dangerous. With an actual study, you could try to measure the curve. Perhaps you would find a large, flat area where changes in speed have small or even negligible effects on safety.

Kind of like the negligible effect on safety that wearing a pillow on an airplane would have.

But we don’t know the answer to the actual interesting question, do we? We don’t know what the exact effect of speed on train safety is. We know that train safety is very different from car safety—I was just looking at some statistics comparing them. But here we are, dodging the real question and complaining that an analogy isn’t realistic enough.

The article glosses over that there were three train-on-train collisions in 1995, and a couple earlier in the 1990s. There was another incident when an IRT sped too fast through Union Square (there's a significant bend in Union Square for the 4/5/6 trains) and derailed.

This may confirm my memory of 2/3 trains hurtling at what I recall as much faster speeds in the early 90s.

Reply on news.ycombinator.com