Caltrain Vibration and Volatility Analysis | Hacker News Reader

Caltrain Vibration and Volatility Analysis

nathanblack.me

76 pointsmmcclure63 comments

He's seeing mostly the suspension problems of the older cars, not the track quality. Caltrain replaced all their track about 15 years ago, going to concrete ties and welded rail.

Caltrain uses a track geometry car twice a year to measure the rail geometry.[1] Those also measure vibration. The Shinkansen system in Japan does that once a week.[2] That's a very good system but extremely high maintenance. The Shinkansen lines have track maintenance every night.

There's a whole cult thing in Japan around "Dr Yellow", the track inspection trains. They're special-purpose 7-car Shinkansen trainsets with automated track inspection gear, and operate at full line speed, so they don't delay traffic.

[1] http://www.caltrain.com/projectsplans/Projects/Caltrain_Capi...

[2] https://en.wikipedia.org/wiki/Doctor_Yellow

Twice a year?!

It's not just Japan. France and Britain check their tracks every 2 weeks.

The British high speed measurement train has a site with pictures of some sensors, statistics ("captures around 10TB of image data every 440 miles") and a video showing more sensors, and an on-train server room.

https://www.networkrail.co.uk/running-the-railway/looking-af...

Well Caltrain doesn't have the wrong kind of snow to contend with.

Edit: In case the snow reference went over anyone's head: https://en.wikipedia.org/wiki/The_wrong_type_of_snow

From your link:

-The train surveys 4,800 miles of track every two weeks.

-The train covers 115,000 miles in a year.

It does not cover the entire system every two weeks.

EDIT: Wikipedia suggests a little over 10,000 miles of track in the UK, so it might be going over it all once a month though I suspect that heavily used high speed lines are covered more frequently than less used lines (which are more comparable to Caltrain).

There are multiple inspection trains, not all of them need to be high speed. The well known "flying banana" is used to cover high speed lines, but for lines with lower top speeds there's also a fleet of loco hauled measurement and inspection trains that cover low speed and branch lines.

This article says the other trains cover routes monthly.

https://rail-record.co.uk/guide-to-network-rail-test-trains/

That's a decent article thanks for posting.

Considering the other problems Caltrian has to deal with not spending much time checking rail straightness above and beyond what's required for the trains to operate seems like a very reasonable allocation of resources.

I think the french do a better job though, the TGV is way smoother than Virgin Trains despite travelling at about twice the average speed. You could do a hand drawn sketch on a TGV, you can't do this on a UK train.

I think this is due to the track being much curvier. Virgin’s trains have to tilt so they can run fast on standard lines. Dedicated high speed lines are built much straighter.

As someone who got stuck in between towns for about an hour while the conductor blamed passengers for the problem – UK rail is mediocre, to put it mildly. The engineer was a bit more constructive and spent his time on the phone with Hitachi. The London Underground is about the only exception here.

Meanwhile tilting trains are designed to tilt so that they can run at faster speeds. They can be designed to ride just fine (e.g. Shinkansen, Amtrak's Acela Express, some of RENFE's AVE lines, etc).

Moving to California gave me a new appreciation for UK rail.

Most of California's rail is low speed and diesel powered. There's almost certainly a higher density of rail service in England, but electrified and/or high speed rail is fairly rare. The infrastructure was so poor around the reign of Thatcher that a lot of the notorious British innovation went into high speed service on low quality tracks.

It's all quite unfortunate as the Brits really had a chance to build out an amazing system (something California hasn't really had the interest in doing). The Underground is pretty great, DLR is at least as dirty and disgusting as BART, and the regional stuff… they don't call it Worst Western Railway for nothing. Electrification is coming at a snail's pace both to California and the UK.

> Caltrain replaced all their track about 15 years ago, going to concrete ties and welded rail.

I live there. I took it to work for 6 years. (I stopped late last year.) The ties are mostly wood. There's some concrete ones in and around some stations, but the vast majority are wood.

Edit: https://www.youtube.com/watch?v=_s9HcflA_nY — there are more than I thought there were, but definitely not all of them.

I’m glad somebody tried to quantify this. Here’s the rocky ride when taking a late Caltrain bullet from north to south (somewhere before Palo Alto): https://imgur.com/gallery/L3LRcld

That's insane! How fast is it going?

I don't think I kept it, but I now understand why it was requested by my friend: I took a video of a thick coin (£1 perhaps, 2.8mm thick) balanced in the train from Malmø to Copenhagen.

I can't remember how long it stayed balanced, at least a couple of minutes, but something like a full wine glass would be fine for the whole journey.

Edit: here's a video of a coin balanced for over 20 minutes on a Chinese HSR train: https://www.youtube.com/watch?v=fumYdO9XknE

Caltrain tops out at 79 mph or about 127 km/h, so most likely 100 km/h or slower.

First I thought "This is going to be a case where a train geek obsesses over a slightly bumpy ride". Then I watched the video and my jaw dropped.

I have no words honestly. 150 year old cog wheel trains in the Alps offer a better ride than this.

Honestly that was atrocious. I can't believe people put up with it. That's a roller coaster ride, not a commuter train. With that sort of movement, you can't eat, drink or do any work. You are lucky to be able to have a nap.

More info: This isn't what the ride is normally like. This was an express train in April, 2019, probably 6pm or 7pm, going south from San Francisco.

Because it's an express the train gets up to some high speeds, and this lateral wobbling starts to really get uncomfortable in the long stretches after Hillsdale or so.

Also, this is the balcony, which amplifies the wobbling. The lower level are much more stable, and I switched to sitting in them shortly after taking this video despite the cramped leg room.

And this was probably the car closest to the engine, like the author of this article said to avoid. I usually sat up there because it was the least crowded.

Caltrain tops out at under 80 mph, that's not really high speed by contemporary standards.

I had wanted to do an analysis like this too -- had even recorded the trace of acceleration using iPhone during rides. But the main thing I thought would be useful would be to overplot the acceleration levels of other train systems in the US and Europe/Asia for comparison.

Caltrain is simply an embarrassment to ride and realize this is what the center of global tech has to put up with. I swear a visitor from any other country with a reasonable rail system would ride Caltrain and think the train was about to shake itself apart. (Yes I realize the old silver gallery cars are especially bad while the Bombardier ones are better)

Even if the gross level of shaking were resolved, Caltrain and most other trains we put up with here are so incredibly loud compared to other more advanced systems. You go to Germany, and even their subways (subways!) are so quiet as to feel self-conscious when you're talking on them. And in the meantime BART -- despite being designed well into the 2nd/3rd generation of US transit systems -- screeches along due to poor rail maintenance.

Funny, I hate the BART screeches so much that I consider Caltrain great in comparison. Guess I'm in blissful ignorance of what is possible. Hopefully the switch to electric trains in a decade will make Caltrain better.

Well actually, I've been told that the problem of rail noise comes from improper maintenance of the rail track, not the motive nature of the cars themselves.

Rail requires periodic grinding to be maintained properly. When you see rail and wheels riding "flat" on each other, that's actually not how it's supposed to be -- that kind of maximum contact actually produces the squealing because the surface area has lots of opportunity to make friction-induced noise.

Supposedly, when you see a rail that's completely shiny on the top surface from being worn down, that's a sign that it hasn't been properly maintained.

A properly ground rail and wheel actually is only interacting on a very small edge of contact -- and this is what produces a quiet ride. A properly maintained rail will be quieter than almost any other mode of transport (noise from the contact surface). It takes active rail grinding though.

So actually, just a change to electric cars will not maintain quiet operations (although in the beginning I'm sure the newness of it will help). BART/Caltrain actually need to conduct the rail maintenance in order to keep it quiet.

Unfortunately, whether it's due to budget priorities being elsewhere, or an indiscriminate customer who puts up with this, unless that aspect changes, we'll be back in the same place. I don't even know whether the conductors -- the ones who would be among the first to notice such issues -- are even trained to understand this.

I encourage anyone to go ride the high speed rails of Germany, France, Italy, China, Japan, Korea, to understand what properly maintained rail is capable of feeling like. Or even the U-bahn in Munich, Frankfurt, too.

Well actually, I've been told that the problem of rail noise comes from improper maintenance of the rail track, not the motive nature of the cars themselves.

Noise on BART is due to a liberal application of NIH. BART literally reinvented the wheel – they used flat profile wheels (nobody else does this) which dramatically increased noise and wear. BART also spaced the tracks too tightly in turns further increasing noise and wear. They made their cars out of aluminum to save weight, creating a bit of an echo chamber (this got worse once they removed the carpeting). Caltrain doesn't have the luxury of lightweight rail cars since they share tracks with freight service.

BART doesn't bother greasing the tracks like nearly every other rail system (almost certainly an artifact of their homebrew train control system). Unfortunately while reinventing everything they opted against decoupling the wheels from each other (a.k.a. using a differential like a car or truck might).

Oh and let's not forget the Oakland Wye, BART's testament to noise, poor service, and political dysfunction.

Simply put BART can't really grind the rail fast enough to keep everything quiet for a meaningful period of time (if they could I assume the wear on the tracks would be astronomical). BART also needs custom rail grinders and whatnot because hey why not throw a custom track gauge into the mix.

So actually, just a change to electric cars will not maintain quiet operations

Caltrain is fairly quiet, at least inside. At the very least the electrification will certainly help with noise while stopped at a station.

Spain is missing from your list, they were instrumental in getting the Pendolino developed and, IIRC, have more high speed track than any other European country.

Oh and let's not forget the Oakland Wye, BART's testament to noise, poor service, and political dysfunction.

Is that worse than the SFO Wye? Samtrans had to create a bus shuttle from Millbrae to SFO because the BART service is so useless for traveling between Caltrain and SFO most times of day.

Sure. SFO is a clusterfuck of BART's own doing, BART's refusal to coordinate with other transit agencies is infuriating (especially at their much ballyhooed multimodal station). Oakland, OTOH, is an example of how little priority the Bay Area places on usable transit. That Wye wouldn't be here but for the one store owner.

but for the one store owner

What's the story behind that??

I'd have to look it up to get more precise details but essentially back when BART was being built out originally they wanted to build through developed land (could've just been the underground rights). They were able to acquire everything except for one parcel. I'm hazy why eminent domain wasn't used (especially since it's been so wildly abused elsewhere in the country in the subsequent decades), but ultimately BART had to build around this property resulting in an absurdly tight curve. The store is gone but the wye remains as a bottleneck – BART trains have to slow to walking speed (ok ok about 18 mph) to go through it.

Edit:

https://sanfrancisco.cbslocal.com/2019/05/07/brake-slamming-...

Rail noise isn't really an issue for Caltrain - but you do have some shaking as well as some noise from the diesel engines. Thus, I expect that electrification and using modern trainsets will create a good experience.

BART on the other hand has horribly loud screeching noises because of its idiosyncratic design choices.

I recently had a conversation at work where I commented how great Caltrain was since I used to commute on BART. My foreign-born coworkers all laughed at me. I look forward to riding some nicer train systems in the future.

Yep - having travelled a lot in Europe it is crazy how quiet some trains there are. Some just have a very light tick tick sound

Yep these are built by Rohr, Alstrom or Bombardier. People who build rolling stock.

The issue is due to insufficient wear leveling, the tracks cause noise (the non independent (fixed) axles cause small wave-like wear on the track surface in track turns which in turn cause the screeching.

The fix is independent wheels, conical wheels or more frequent leveling of tracks.

Caltrain will soon get modern electrical trains [1] which have been in use for 8 years in Switzerland [2] and which will solve the vibration issues.

[1] https://calmod.org/electric-trains/

[2] https://en.wikipedia.org/wiki/Stadler_KISS

It seems like this isn't going to be a reality for another half-a-decade at least. (2+ years of constructions and 2+ years of productionization for consumer traffic).

Unfortunately since there's only two tracks, one north, one south, it's kind of a farce. There isn't enough room for trains to over take.

Thanks for doing this analysis!

After riding other transit in the US (NJ Transit, LIRR, Amtrak, BART, misc Subways), high speed rail in China and Japan, and trains in Europe; Caltrain stands out to me as the most jerky train I've been on. Often times high speed rail in East Asian countries are so smooth that you can balance a coin on its edge without falling. Comparing apples to apples, even NJ Transit and LIRR are noticeably smoother than Caltrain. It would be interesting to see what the difference is between systems.

> Looking at data from a few different trips, Caltrain does a good job subjecting passengers to no more than 1g of roughness in any direction.

This at first to me seemed like a low bar (relative to other trains out there), but then I forgot about the sudden jerks and direction changes the SF Muni and NY Subway often subject passengers. 1g in any direction I suppose is good for subways, but still poor for trains overall.

Unfortunately I doubt Caltrain's current Electrification project won't solve any of these issues, given that it seems like more issues with the track, not with the power source. Perhaps the new trains might help?

Not as hot in the summer for east coast commuter rail.

Much of the side-to-side jerking is a combination of the poor suspension on the trucks (bogies) of the older Gallery train cars and fast running over track switches. The Gallery cars are the ones that look like steel toasters.

If you take a ride in one of the "newer" Bombardier cars (the octagonal ones) the ride is much smoother and the bumps are almost imperceptible with little swaying. I'd imagine the new trains post-electrification will also be much better on the suspension front.

> I'd imagine the new trains post-electrification will also be much better on the suspension front.

So right now this is still Diesel?

Yes. It’ll be a couple years before Caltrain has electric trains.

Yep. Towns along the peninsula (looking at you Atherton) are fighting electrification (as well as grade separation) tooth and nail.

Why? What's the disadvantage for them?

Just the installation cost?

Mostly they don't want to mingle with the hoi polloi.

Atherton (not Belmont – all those shitty little peninsula towns blend together in my mind) sued because they didn't want trees removed to make way for overhead wires (the claim was that "cleaner" diesel would be better than electric)[0]. Then they sued because the poles were going to be 10 ft higher than promised[1]. Another Atherton resident[2] sued because he wanted Union Pacific to explicitly approve everything beforehand. More dreck[3].

Caltrain's getting sued left and right over the electrification, starved for money by BART, and generally ignored by politicians. If you're ever curious about why Bay Area transit is as bad as it is, Caltrain is a great case study.

0: https://en.wikipedia.org/wiki/Electrification_of_Caltrain#La...

1: https://www.almanacnews.com/news/2017/12/21/atherton-battlin...

2: http://www.cahsrblog.com/2009/08/another-peninsula-nimby-law...

3: https://caltrain-hsr.blogspot.com/2009/08/atherton-et-al-law...

Not hard to see what their real intent was.

When some have the freedom of protecting their private land, others end up with the freedom of riding diesel only, underfunded trains. The liberty of some seems greater than others’.

I wonder a bit though as advocating for diesels service seems a bit like they're shooting themselves in the foot, at least in terms of air quality.

Belmont at least has some density and Main Street. Atherton’s density is 1/4 of Belmont’s and it occupied the very center of SV.

Atherton 1,438.97/sq mi (555.53/km2)

Belmont 5,843.98/sq mi (2,256.45/km2)

San Mateo 8,570.40/sq mi (3,309.14/km2)

To make matters even more frustrating, Atherton starved their Caltrain station to death:

https://www.almanacnews.com/news/2020/01/17/atherton-signs-o...

I imagine there's also quite a bit of hunting / side-to-side movement as well. BART tried to solve this with flat wheels, and after forty years of complaints came to the conclusion that flat wheels wear the track down more and cause significant noise. They've since cut a conical profile (like most rail operators) into the wheels under the guise of reducing noise. I can't hear much difference but the side-to-side motion is quite a bit more dramatic than I remember.

It's been almost a decade since I've ridden Caltrain with any regularity but I don't remember being particularly enamored with how the Bombardier cars rode. If anything they rode firmer than the Nippon Sharyo cars over the bumps but had less lateral movement as a consequence.

I remember there being quite a bit of variation (there are three distinct batches of them after all) – something you used to see with Muni buses. The German ones (Neoplan) touted their complex (independent) front suspension, and there was a huge variation in ride comfort with some being downright scary. The Hungarian (Ikarus) ones with their simpler solid axle up front seemed more consistently comfortable.

Electrification will involve entirely new train sets, so it’ll go a very long ways towards fixing things. Electrified engines aren’t going to be pulling the older cars.

Soem comments on both the article, and other comments here:

- It's an interesting assessment, but without know the details of how he did the measurement, take care in interpreting the results. If he was using a smart-phone just lying on s seat of floor, the data are almost certainly (mostly) meaningless. If he was using a professional triaxial accelerometer that was rigidly or semi-rigidly attached to the floor, the data are more representative.

- With regards to comments about the data being related to track conditions. Passenger trains have two sets of suspension - the primary suspension between the axles and trucks (bogies for our European/Asian friends) and the secondary suspension between the truck and the passenger car. The secondary suspension is typically very soft for ride-quality purposes and is mostly isolated from the track. The vertical vibration levels measured here are almost entirely a result of the ride-quality decisions made by the manufacturer and agency engineers, and less so about the track condition.

- As pointed out by nhf and inferiorhuman, the side-to-side jerkiness (lateral vibfration) is likely related to track condition and is some combination of track alignment problems and wheel hunting (there the wheels on the solid axle bounce sideways back-and-forth between the parallel rails).

- Electrifying the line will have no affect on the cabin vibration unless Caltrains elects to purchase entirely new passenger cars with different secondary suspension design goals. If the new electric locomotives are pulling/pushing the same passenger coaches, the vibration will be the same.

- From personal experience I can tell you that the TransRapid TR08 (German system, used in China) can gave similar subjective levels of jerkiness when the guideway configuration has a sudden change.

Caltrain is buying modern EMU stock where traction is distributed across the passenger cars.

From the synopsis:

> It's important to use dynamic measures in a dynamic system. I often see the mistake of using a static value for mean or kurtosis, unaware of how it changes over time.

This is both important to note, and unavoidable. You can account for changes in volatility in pure data analysis, but when developing mathematical models of time series, data stationarity has to be assumed for closed form solutions. Even if the volatility itself followed its own model (say garch could model it accurately), it can be iteratively reduced to a stationary series (I'm not sure if there's proof of this, but that's how most time series courses teach it - take first diff, second diff, log etc).

Markets are a similar level of complexity, but to compare anything to a reflexive system such as a market is kinda skirting the main issue of markets - reflexivity. You cannot find a closed form solution for it using current mathematics. Finance developed a distribution based mathematics called stochastic calculus (used in other fields, but it's the backbone of finance) to deal with this, but even there, the assumptions are highly restrictive compared to real world conditions.

This article is somewhat interesting but it appears to be a not so subtle advertisement for a trading platform.

Really curious where you're getting that impression.

> I like rhythm and vibration and I trade financial market volatility at sha.capital.

This is the only mention of anything associated with his work, and it's not even hyperlinked. It's a fund at that, not a platform, and he doesn't even make an attempt at any kind of solicitation in the post. If this is an attempt at advertising it's a pretty poor one.

What would be interesting next would be to see what the track geometry is like in the area of & around the worst events.

Something like what you’d get from this: https://youtu.be/Ite0l4uJ_fM?t=3m55s

This is fun. Interesting analysis. A few quick thoughts:

Why the focus on Jerk/Volatility instead of Acceleration/Force? The energy imparted to the passengers is proportional to the forces. Is there any reason to suppose that Jerk is more important to the experience of discomfort (or mechanical wear for that matter)?

The analysis notes that the distribution is not normal, but it does not explore the what the underlying distribution might be instead. Given that the train's velocity is constantly accelerating/decelerating it seems unlikely that it would be a single distribution. However it may be the case that the observed variations are the result of a superposition of normal distributions. A Fourier transform could maybe helpful to explore these aspects.

Acceleration/Force is easily compensated for, if its constant. For example, taking a turn or coming to a gradual stop is completely comfortable. Only the change is uncomfortable.

Jerk is what's unpleasant and annoying. In machine design it's something that is carefully dealt with since it has a strong effect on wear. See cam design.

I'm interested in the controls for this analysis, which I didn't notice in the article. Might've missed them.

I'm mainly talking about how the device was mounted - because if it's simply kept on person, a lot of this vibration is muffled, right?

I've played with the thought of doing something like this to create a tool for assisting (and perhaps evaluating) drivers of buses:

* make a device that logs acceleration in various directions and performs suitable statistical analysis in real time. * combine it with passenger evaluations of ride behavior (tell users to press a button when the ride is uncomfortable) to train a neural net to evaluate ride smoothness * install devices in all buses, with a display placed in front of the driver that shows current ride comfort and warns if things get too problematic

Just something I noticed [there is a plate near the doors], cars were built in 1984 and refurbished by Alstom few years ago.

Fun read - great article for getting a better intuition on real world data analysis of dynamic systems (especially the part showing how bad conflating mean and median is).

Only gripe is it’s not an interactive notebook of some sort so others could play with the data.

The deviation from a normal distribution is really interesting!

Reply on news.ycombinator.com