Steam speed records including City of Truro and Mallard

Now that's what I call thread drift!

 

In which case he has my sympathy.

 

Yes, of course there are a couple of finer points! But the relative chimney length argument is common to double chimneys, Kylchaps, Lemaitres, Giesls and Lempors and even to the Russian early example of 1900!
Kind regards,
Jos

 

I would challenge anyone to come up with an alternative practical design with effectively eight chimneys!

 

Fix It Again Tomorrow

 

Divide 127 by 50, gives the number of centimetres in an inch, definitely a useful number

Michael Rowe

 

Mr Andrews is referring to the way the dynamometer car records the seconds/distance/etc and having studied what I can from the records available, and primarily the associated documentation at the national archives, I do not believe his interpretation is correct, particularly as - again - he’s only using half the data actually available within the dynamometer roll.

 

Are you able to describe exactly what your analysis is - I'm not asking for your final answer on speed, or to display the whole roll - but at least describe what the additional data are and how you are treating them. What are all the data sets recorded on the roll, and what is their source? (continuous pen trace derived form measuring wheel, point mark from observer button click, hand written annotation etc etc).

In particular - are you confident that the additional data you see is genuinely independent, and isn't just the same data recorded twice? The reason for asking is that earlier in this thread you plotted a graph that had two traces on it - but actually they were simply the same data in mph and km/h, in other words you had plotted two data sets but really there was only one. It would seem surprising to me if there was an entirely independent set of data "hiding in plain sight" on the dynamometer roll that had been missed by every previous person looking at it - not least by Gresley himself and his technical assistants.

As I understand the argument in Mr Andrew's paper, he had isolated a periodic variation in the recorded speed, which - given its regularity over distance - must almost certainly have been some minor imperfection in the dynamometer car machinery. When you subtracted that cyclic variation from the recorded speeds, you got a much smoother curve. The 126mph claim then seemed to rest upon having taken a single point at the very peak of the cyclic variation. His analysis is both better and worse from the point of view of the loco: "worse" in the sense that the likely top speed was nearer 124 than 126 mph; but "better" in that that speed was sustained for a longer distance. Apart from anything else, the speed curve he plotted seemed much more likely with regard real-world vehicle dynamics: 400 ton trains do not suddenly gain 1mph in a matter of seconds and then lose it again just as quickly.

I think his analysis is quite convincing, and I'm struggling to understand what additional independent data you have that refutes it.

Tom

 

I mean Tom, I’ve provided several screenshots of the table. Look at it and read the columns.

 

A calculation doesn't create new data. If you take two numbers and divide one by the other, the result doesn't constitute new data, because it is not independent of the original two. (If you have distance and time, you can calculate all sorts of interesting things, notably speed and acceleration. But those calculated numbers aren't themselves new data: they are simply derivations of the original data. The only genuine data you have remains the distance and time).

Your claim is that Mr Andrews is "only using half the data actually available within the dynamometer roll." So my question is - what data has he missed? If you are saying "I'm taking the same source data but presenting a different analysis" - well fine, but that isn't what you are saying. Your claim is that you have new data that have been previously overlooked: my question is what is the nature of those new data.

Tom

 

Tom. I'm using every quarter mile measurement from the roll, taking the time between each quarter mile as represented on the roll.

Mr Andrews only used the mile and half mile points that the LNER team used.

I have more plot points because I am using all of the data available.

Mr Andrews has then made a claim of the mechanism. I looked into this before and the little up tick is designed into the mechanism to provide a point of reference for consistent calculations. So in effect, yes, he was wrong.

I have never claimed I have "new data". I have the original data, which I can now calculate more values from.

I hope that clarifies what I am saying and what I am doing.

 

So the blips ( since they are merely markers ) should be ignored. Is that
not what David Andrews curve does ?

Michael Rowe

 

I'm getting a tad confused with these traces. In my naivety I was thinking that the dyno roll was driven directly from the fifth wheel because every similar plotter I've ever used has had distance as the 'X' base by default with a time frequency provided by a regular 'tick ' to indicate time, generally every second. The 'Y' pen is usually speed but is a derived value and can be checked against the speed and distance information. I'm now getting vibes that the dynamometer roll doesn't follow this form and is merely a roll of paper travelling at an undefined and perhaps variable speed on which distance and time plots are simply indicated by pen 'ticks'. That's open to more error as there isn't a common zero to measure against. A screen shot of at least some of the trace would be advantageous here, even if it wasn't the critical moment.
I'll just sit back with the popcorn.

 

I think we need to take a couple of steps back here.

Steve, you are correct, the dynamometer roll does run off the fifth wheel. The revolutions of the wheel run the paper through the measuring pens, which are set to record things at specific rates.

Distance is measured by the speed of the wheels.

The distance between each quarter mile is given by the pen, and the dynamometer is set up to identify each quarter, half, three quarter and complete mile by way of a specific sine wave.

Time is in seconds and is a consistent standard size of sine wave on the roll.

Steam application is also recorded, as is DBHP.

Dr Andrews paper postulated that there was an error in the sine wave and that apparently was an issue with the gears in the machine.

I can confirm that’s not the case, it is designed to give a marker at the top and bottom of each square sine wave to record from.

We know this because some copies of part of the dynamometers operating instructions have survived in several archives. So we can eliminate that error. He didn’t need to adjust for that - the uptick and downtip is just a marker for calculations.

The bit I am interested in is that the speeds were only calculated by the LNER team to half and full miles, but the dynamometer roll gives each quarter mile. By not using this part of their own data, the LNER speed curve looks more abrupt than it likely actually is.

What I am doing differently is measuring the time recorded between each quarter mile. That means I already have more data points on my graph than both the LNER team and Dr Andrews.

I have already done a lot of work on it and the current prognosis is that Dr Andrews and I both agree that Mallard sustained 124mph+ for a longer period of time. However, as I am able to give instantaneous speed at more points, I may yet have more to reveal as I get the data inputted.

My gut feeling is that the more interesting thing to analyse is the acceleration up stoke bank.

 

No, because he used the plotted points by the LNER team and also used the 5 second intervals they also plotted, as the basis for his attempt, which I feel is an error in approach. The five second intervals take an average speed but miss the rest of the potential data plots on there.

 

Sorry Simon - but what's a "square sine wave"? It doesn't exist.

A sine wave is a curve. In particular, it is the curve you would get if you marked a point on the circumference of a wheel as it rolled through one revolution and then plotted the height of that point above an arbitrary baseline (for example, rail level) against the linear distance travelled by the wheel centre.

The square form timing markers are the analogue representation on a chart of what is a binary (on / off) electrical signal generated by the timing circuit. That goes on and off every 1/2 second. Because there is a small hysteresis in the response of the recording pen, you don't get a perfect square with right angle corners, but instead get very slight slope to the rise and fall, with a flat middle part. But in this context, that is just illustrative that nothing is absolutely perfect: it doesn't fundamentally affect the positions of the 1/2 second markers. (The blip gives a useful measuring point, but tempered by the fact that the line has a non-zero thickness). But that is of little consequence unless you want to start measuring how far the loco travelled in each individual 1/2 second, which would have a large error bar. Stretch the timing intervals out over which you calculate average speed, and you get more precision. (Not more accuracy, but more precision).

In the Andrews paper, he identifies a sinusoidal curve in the apparent rise and fall of the speed: "When this sine wave was subtracted from Mallard’s speed curve the constantly rising and falling trace disappeared". If you see a sine wave in any trace, it is very indicative that it is related to some form of rotary motion. In this particular case, since the rise and fall is over the course of 1 mile, it implies that whatever is causing it is something that takes one mile to complete a full revolution - it will be one of the gear wheels in mechanism that rotates once every mile. He is not talking about the square form timing curve; he is talking about an analysis of the apparent instantaneous speed which waves up and down (on top of its "true" value) in a regular cyclic fashion.

Coming back to data: the speeds written by hand on the chart are not data: they represent after-the-fact analysis of the underlying data by the LNER technical assistants. The only data relevant to the speed that I can see (based on what has been said about the trace) is represented by the square form timing marks, and how far they are from a datum point on the roll. At 120mph each mile takes 30 seconds, so you get 60 square waves. Since the chart is moving at 24 inches per mile, you get 60 waves in 24 inches, i.e. each one is about 0.4 inches (around 10mm) long. When the train is going slower, they are shorter - at 60mph they would be 0.2 inches long.

If you want the absolute data from which all analysis is derived, and you have access to the roll itself, then measure the starting point of each wave relative to a datum point on the chart. (Measure each one back to the datum rather than from the previous wave, so you don't get cumulative error). That will give you a distance travelled at half second intervals over a period of many miles either side of the record moment. You'll have thousands of data points on which to do your analysis. From that you could calculate the instantaneous speed every half a second: what you will get will be incredibly noisy (i.e. one number might say 140 mph and the next calculation 1/2 second later 100); but from that you draw a smooth curve with justification for the smoothing functions you applied.

I assume, incidentally, that that is basically what the LNER did, except that the tedium of doing individual calulations by hand meant they took five second intervals, rather than half second, cutting the number of data points - and calculations to be done - by 90%. That also smooths the curve of much, but not all, of its underlying noise. For their purposes it was no doubt “good enough” since, in the end, the last mph of calculation is hardly here or there - they seemed quite happy with 125mph. What Andrews then did was essentially a form of analysis that was able to isolate the harmonic (sinusoidal) variation in apparent speed introduced by the measuring equipment.

Tom

 

Maybe before commenting, you should make yourself aware of what you are commenting on? Post #476 should give you a clue.

 

Hi Tom - perhaps I’d better just give up and hand you the roll then?

 

As I see the Andrews curves there are two sine distortions at work;once per mile and four times per mile and phaselocked but not quite symmetric.
If someone living near York can be allowed to see and describe the dynocar gearing between fifth wheel and paper drive it can be estimated where fault can be.
How precise consistent is the clock cirkuit for the time marker pen drive?
Same person can do us all a favour and ask where the three rows of speed estimates stem from?Nobody here knows evidently and they are quite fundamental to the problems we try to solve.

Mallard by Hermod posted Dec 5, 2023 at 10:03 AM

 

Tom has written a cogent summary of the data anomalies and the probable cause. His analysis is one with which I entirely agree. If, many years ago, you had been a student whom I was supervising, I should have requested that you examine Mr. Andrews' paper and Tom's comments thoroughly and if you think them wrong, explain why equally cogently.