Steam speed records including City of Truro and Mallard

I was expecting responses, but I am taken aback a little by the strength of the responses.

For clarity, no, I do not have a “number in mind” for the “speed record”.

I have always maintained that I thought there was more to the data than is shown in the full run and it is my intention to plot the whole of the run.

Much of the writing on the run hasn’t actually used all of the data: there are far more points to plot, actually, almost double than has been shown before by simply examining the chart and calculating the speeds in a different manner. That’s actually more of the driver for doing this exercise: I don’t believe previous writings on the speed run have looked at the roll carefully enough and my approach is different and will produce likely different results.

I’m more interested in showing the whole run as an exercise in overall performance including acceleration (and again, I have consistently said in this thread that the run UP stoke bank looks more interesting than the run downhill) together with being able to show comparatively what everyone’s interpretations look like. That in itself is a useful exercise as it allows those reading to make their own minds up but to do so with all of the data.

The error bar discussion was more about opening up the floor for different points of view to emerge, I have had some preliminary discussion elsewhere and I have in mind what I will likely achieve with it.

Overall, I am a bit disappointed coming back from work to see quite a negative reaction to what I am trying to do with the data.

Yes, I have an LNER bias and I would love it if it showed 126mph - but if it doesn’t, tough! Evidence based approaches need to be evidence based approaches. If this doesn’t show 126mph, so be it, I will revise my views.

 

I think there are two things we're struggling with here. Firstly, at a statistical level, once you introduce error bars that equate to more than what, 0.5mph? I don't see how you can actually prove anything new with any level certainty. As Tom says, you have to go back to your data collecting devices and get their individual precision and accuracy, and add them all up, to get your error bars. Sadly I chucked out all my Physics A-level notes, but it was drummed into me by my physics teacher. She used to tear up pages of work if your final figures had more significant figures than the input data. There was no point in calculating the speed of the mass we'd just dropped to anything more than 2 s.f. as our reaction times in timing the drops by eye weren't good enough to do it any better.

On a big picture level though, whether Mallard did 125.5 or 126.0 (or even 126.0000!) how does that change our understanding of the development of high speed rail, which is supposedly the ultimate aim? It's miles off the excellent statistical analysis you did for Thompson, not in terms of quality, but in usefulness, where it genuinely gave us a new insight.

I think fundamentally, aside from an interesting theoretical number-crunching exercise, we're struggling to get excited about the output because it's not going to mean anything significant either way, for statistical and big picture reasons. Sorry!

 

Thank You for a very clear letter that send me thinking.
Most of the 126 believers are dead in a few years and few young ones left will be quite about it for not being judged nerds or in needs of physcological assistance.
It is I think the underlying Hero making of brave CMEs that irritated me.What people worship is not my buissness and I shall restrain myself in future.(If I can)

 

All the while the 126mph plaque is on the side of Mallard's casing, it will be there for all to see and no amount of discussion in this thread - especially by people who weren't even born back in 1938 - will change that.

 

No, It is because of fundamental fluid dynamics! I have explained it before in the 6023 and exhaust discussions, but here it is again:
Already in 1863 the frenchmen Nozo and Geoffroy tested locomotives with 4 and more scaled chimneys. They found hardly any difference with a single exhaust system. Why this was so was not well understood, but the american Buckingham wrote in 1913
an article with an explanation: models give model performance! His article was fundamental for model testing, but for us now is important that a single exhaust system can be replaced by 4 model systems, which have 1/2 scale. However, we use 4 scaled orifices, but not the 4 model chimneys, the original is used. Compared to the models this chimney has the double length and its behaviour is quite like the four models, but its length adds some extra vigour in its working. A tapered chimney is called a diffuser in
fluid dynamics and longer diffusers work better in their conversion from pressure into velocity.
It is why the single orifice did not work in the 6023 with its shortened chimney but the replacement by a fourfold orifice got her functioning. Please note that this explanation can be found in different chapters in modern day textbooks!
Kind regards
Jos

 

May I suggest

1. @S.A.C. Martin promises to come up with something interesting in his analysis of the Dynamometer roll , but
2. @Jamessquared rightly points out that the significance of this is limited and there is more interesting and fundamental stuff to research and analyse

The challenge is to balance the two

 

Yes, exactly. With any time and distance recording we don't measure speed, we calculate speed from time to travel a distance. Both the time measurement and the distance measurement will be of limited accuracy so we put them together and calculate a single number speed, but what we actually have is a range of speeds. So to get error bars we have to establish what the longest and shortest distance the distance run could have been, and what the longest and shortest time. Rous Marten's 1/5 second stopwatch provides an excellent example. When his stopwatch said 8.8 seconds it was really anything between 8.7 and 8.9 seconds, so that's our first error bar - about 1 and a half percent plus/minus or 101.12 to 103.45, and an actual 8.8 second 1/4 mile could be anywhere between those numbers. So quoting the speed at 102.3 is really a nonsense. 102 would be better, 101-103 better yet, and 99-105 even more reliable. We might also consider that he might have clicked his stopwatch just to early or just to late, and got 8.8 when it should have been 8.6 or 9.0, so for our second error bar we get anything between 8.5 and 9.1 seconds - say 3.5% plus minus. We also need to consider how accurately the mile posts were laid out, although in practice this will have been rather small against the very limited accuracy of the stopwatch and we can pretty much ignore it. One may then do the same for all Rous Marten's figures and draw a chart. When I did it I came to the conclusion that a 8.8 quarter mile was feasible against the other figures, but the real speed would have been close to the bottom of the range - probably nearer 101 than 102. Of course those who don't believe Rous Marten recorded his times correctly can discard all that.

Now when it comes to Mallards dyno trace we are dealing with something much more precise than 1/5 of a second, but it will still have its limits. There are a whole bunch of things we can do though. The roll contains both distance measured by the wheel and mileposts. We can use those to cross check each other. A really keen option would be to take a trip up and down the LNER mainline with a couple of GPS recording devices, and use those to calibrate against ALL the geographical points mentioned on the trace. That will give more information about how well the mileposts and geographical points match up to modern GPS traces (plus get much more accurate measurement of the gradients as they are today, which would be useful too). So if we line up GPS points, chart geographical points and chart mileage run we have a pretty good idea of how accurate the roll is for location. And what we'll get, doubtless is a decent confidence that distance run is accurate to within 1%, 0.5%, whatever, and maybe a correction if we find that say 100.5 miles on the trace actually represents 99.7 miles by GPS.

Then we have to do a similar but less precise exercise for time recording, which will also give us an accuracy, and finally combine the two to get the error bars for the recording. Its a shedload of work, don't underestimate!

An important thing to consider is that once we get down to the limits any number within the range is valid, so the data has a lot of noise. The longer the distance/time we measure over the less important the noise is, so that's why we average over a reasonable period. The longer the period the more accurate the calculation will be, but also the less extreme it will be. A single 8.8 quarter mile might be within a 17.8 half mile or a 37.2 mile. Each figure is more accurate than the one before, but also nearer to the mean.

 

That’s what I was trying to push towards, actually! The dyno roll does allow us to get the data represented to an interesting degree of certainty due to how it records the time output on the roll. Two decimal points is used by the LNER team throughout. I would be surprised if by using two decimal figures myself, that in and of itself would be a problem.

The reason i asked the question was to try and generate a discussion on the presumed or anticipated potential errors in measurement. How far do we go? Tom is right to point to the dynamometer car as the issue, how accurately we feel it measures distance/time/etc is part of the question. I am doing it as an exercise in curiosity to see what more we can learn.

Because, again, it’s not the downhill bit I’m most interested in. It’s the acceleration up the bank. Mallard represents that final development in high speed steam locomotive design prior to the Second World War and understanding the locomotive’s run (throughout its high speed record attempt) tells us a lot

I’m only a month into a 4-7 year PhD, give me a break…!

Fair enough.

 

All true Jim - but the data set itself also matters. The Dyno roll shows waaaaaaay more potential plot points on it than has been previously reported and it is this I am interested in. I suspect the additional plot points will help to show the noise you are talking about - making it clearer to us if 126mph was achieved outright or was 125 a safe claim, and 126mph erring on the liberal with the data?

But the big thing for me is Mallard’s acceleration and sustained speed, from which with more data points we can see a more accurate picture of. That might help us inform other matters too.

 

That's because they are using GPS badly. The instantaneous speed shown on a GPS device is nearly useless. What you have to do is log all the data and analyse it. You can then discard all the points that are obvious GPS glitches (position not over the track is obvious, but also rapid height changes may indicate issues).
A typical GPS trace might look something like this. This is actually from my raceboat, and you can see that it was moving around in roughly a 10 metre circle. Now here's the secret. It wasn't. The boat was firmly tied up against a jetty, and the most it could possibly have moved was about 150mm in any direction. All the rest is GPS inaccuracy, so you can figure that with my 2007 phone and the satellites that were up at the time the best you could hope for was plus minus about 5 metres. You can also see that there are lots of traces in a bunch, with occasional bursts of movement. I expect this reflects changing between satellites.

So given a log of a train run, a speed between any two successive points, which is what the instant readout typically gives, is of very little value. But if you take two GPS points half a mile apart then the the error is still +- 5 metres, so the accuracy is way way higher.

 

This is where you have to understand the error bars. If the measurement tool you have delivers two decimal points record them. But understand the last one, maybe the last two or more may not be reliable.
I am reminded of when, at primary school, I proudly calculated pi to about 30 decimal points and showed my teacher. He kindly pointed out how my number differed from the true one and explained what I was really doing was finding out how accurately I could measure a circle!

 

The altitude can also be glitchy (if you wanted to use it to measure a gradient profile, for example).

This is an altitude trace I got while out running. Apparently, over a distance travelled of 60 metres between 2.60 and 2.66km, my altitude went from +49m to -84m.



Given where I was, I'm reasonably certain I wasn't at -84m (I would have got very wet if so!)

What actually happened was that someone stopped me to ask directions. I paused my watch, spoke to them for a few minutes, then resumed. I've got conflicting info about how my phone calculates altitude, so I can't exactly explain the discrepancy, but clearly the base reference changed in the few minutes I was paused so when it resumed it locked on to a different value. I've noticed it happen a few times.

(FWIW, according to OS maps, at that point I was between the 50m and 55m contours - probably about 52m or 53m would be right. So 49m was moderately wrong but the relative up and down shown are fairly close to reality: the -84m is demonstrably wrong.)

Tom

 

It's a problem if you want to give that level of precision to a very fleeting event, i.e. a peak speed recorded over a few seconds.

If the LNER ran 400 miles to Edinburgh in 6 hours, then you can probably give an average speed to four significant figures quite legitimately. A timing error of 1 second (assuming your timekeeper sat on the train with the same watch throughout) in 6 hours is about 1 part in 20,000. I suspect the distance accuracy is considerably better than 1/4 mile: getting the distance wrong by 350 yards feels quite sloppy but would still be accurate to 1 part in 2000 - the distance is probably more precise than that. So the 400 miles in 6 hours is 66.67mph and you are probably fairly confident to that second decimal place.

But try the same thing over 5 seconds and to get the same level of precision, you need to get your timing and distance accurate to the same 1 part in 2000 level - so your timing needs to be +/- a couple of milliseconds; and your distances need to be +/- 8 inches in every quarter of a mile - which is the thickness of a mile post. The dynamometer doesn't support that level of precision - so your final figure needs a similar lack of stated precision.

The longer the length you give a figure for, the more precision you can offer. You can have more confidence in e.g. 5 miles at 120 mph than you can have in 5 seconds at 126. (A 0.2 seconds timing error over 5 seconds gives you. +/- of 4%, so limits the precision to at best +/- 5mph. The same 0.2 seconds timing error over the 2.5 minutes it would take to cover 5 miles give you a timing error of about 1 part in 750, so might give you confidence to a first decimal of a mile per hour - certainly to the nearest mph.)

Tom

 

Some pages back I asked if there was a before and after calibration check on the dynamometer car instrumentation and, if so, did it exist? This is a fundamental and would help a lot in the assessment of what happened. It’s not rocket science none to move the dynamometer car over a measured distance and check the roll movement against it and the time trace over a given time. If they remain the same at start and end then the results should be valid. When I used to do instrumented testing it was the first and last thing you did.
I don’t know too much about the car but I’m guessing that the chart was mechanically driven from the fifth wheel and the time marks from a chronometer so the likelihood of variance is limited. I accept the argument about the greater the length and duration, the less the error but I do think with what we have it should be possible to get to at least a decimal point if my understanding of the roll and pen drives are right.

 

In the case of the rebuilt 3500 Class, that's clearly not true. As a consequence of their improved performance, the timing between Angouleme and Poitiers was changed, together with the time allowed for Paris to Bordeaux, on trains that could exceed 800 tons. These timings were only used once there were sufficient rebuilt 3500 Class, which became the 3700 Class.

Chapelon calculated figures of 25% for the streamlining of exhaust and steam passages, 10% for a 100degC increase in superheat and 25% for the adoption of thee Kylchap. So the Kylchap would have accounted for slightly more than 40% of the increase in power. Testing of the locomotive proved Chapelon's calculations to be correct.

That's why compounding in France was limited to express passenger and fast mixed traffic locomotives. Most other locomotives were simple expansion.

 

Right. But the dyno roll has an automatic distance recording. It’s not dependent on a timekeeper, it’s automatic. We can be confident of a consistently reasonable recording to that end, same thing for time in seconds.

Yes the LNER did calibrations on the dynamometer car. I am following up some leads on private collections which may have a full copy (I have found to date only a partial description of the procedures).

The LNER team went to two decimal places and considered that acceptable for speed. You are correct regarding how the dynometer car works, however the distance between quarter mile marks on the roll gets smaller as the speed increases. Only time, measured in seconds, does not vary on the dyno roll.

This is actually an advantage because we have two known measurements - quarter miles and time. Which means we can calculate using the usual distance time equations to get instantaneous speed at each quarter mile (which is not something which has been done to date, as it happens).

 

Well no, the LNER recorded two decimal places. But when they reported the speed to the public they went no further than the integer. An important distinction.
The time ticks will vary. everything varies. A useful exercise would be to measure a large number of them as precisely as possible and see how much they vary.

 

Always dangerous to assume! What's the saying? To assume is to make an ass of u and me? Before making such a statement, it might have been advisable to make yourself familiar with the actual case.

 

I was referring to what's on the dyno roll, not the PR by the LNER. The team on board the train, and those who analysed the roll afterwards, went to two decimal places for almost all of their calculations.

Interestingly, at odds with the rest of the roll, the speed in MPH gives miles and 1/2 miles for the 5 second interval recordings.

Exactly what I have been doing. The variations are negligible: the equipment was clearly in good order. We can consider the recorded seconds as accurate.