Great Western Coffee Shop

From the BBC (http://www.bbc.co.uk/news/world-europe-34822666):

Later reports say "at least" 10 people have died, from 49 on board. This was a test run on a newly constructed line between Paris and Strasbourg.

Very sad  :'(

I hope that this is a "normal" railway accident and not related to yesterdays terrorist attacks in Paris.

The fact that some form of test or trial was underway, rather than a scheduled passenger service, does suggest an accident perhaps whilst travelling at a greater speed than would be permitted normally.

Looking at the aerial picture in this report (https://www.rt.com/news/322054-tgv-train-derails-eckwersheim/), an explanation in terms of "too fast for the curve" looks all to plausible. Whether that means a test run intentionally at higher speed than will be permitted later is another matter; the investigation will tell us that.

The other factor is that the front of the train is reported to have struck the bridge parapet, explaining the violence of the impact and how far the carriages have been thrown sideways.

Correction: other reports say the train was going toward Strasbourg, making this a right-hand bend - so most of the train kept going straight on and has jumped across the canal.

Inevitably this was not a major news item - there's just too much news today.

Indeed: a very sad day in France today.  :(

As one of our members now says in his footer text: "Je suis un Parisien."  :-X

An update, from the BBC (http://www.bbc.co.uk/news/world-europe-34822666):

Latest TV report says that in addition to 10 confirmed deaths, there are 12 seriously injured and 5 still missing when the underwater search was suspended for the night. There were 49 aboard, including the families of some railway staff - this run was seen as marking the end of the testing phase. However, it was described doing "elevated speed" tests, and 350 km/hr was quoted but I still find it hard to believe that's possible for this curve (and it came from S^gol^ne Royale, who has a record of mis-speaking).

An update, from the BBC (http://www.bbc.co.uk/news/world-europe-34827049):

I read a French report last night, which was more detailed. That suggested that the test crew had taken family members on what was the final test run. It did not say whether or not SNCF had sanctioned this.

The test phase using this instrumented TGV-Est train started at the end of September and runs to the end of the year, so is only about half-way through. This first part is described as proving the line at increasing speed, from 160 km/hr up to the full train and line speed of 320 km/hr and then 10% higher (352 km/hr; hence the figure of 350 km/hr that has been quoted). The site of the derailment is on a curve with a radius of about 1.3 km and just over 1 km from where it joins the existing "classic" line for the run into Strasbourg.

The five "missing" persons are now believed not to have been on board, so the casualty number is as in that last report: 11 killed and 5 still critically ill in hospital. One of those killed was a child, and their presence has still not been explained. Other reports suggest this kind of "treat" was common practice, though unofficial. If the tests have in fact reached full speed, that might seem to be the end of testing the track for faults - leaving things like the signalling and power supplies, and running over every piece of trackwork in every direction. Incidentally, on the LGV trains run on the left, so this run was in reverse. (Trains in Alsace run on the right, so the junction just ahead has to accommodate this switch.)

The investigators (SNCF, BEA-TT and gendarmerie) have recovered the train's data recorders, though presumably the signalling system will know its position and speed. The line is equipped with ETCS/ERTMS, but may not be using it at this stage; it also has the older TVM 430 system. BEA-TT do not usually give the same kind of full statement of facts at the start of an inquiry that RAIB does. However, for serious accidents like this one, SNCF will probably do something similar in a few days, and that should give the actual speed and the limit for this curve.

According to le Parisien (http://www.leparisien.fr/faits-divers/deraillement-de-la-rame-d-essai-du-tgv-est-le-bilan-s-alourdit-15-11-2015-5279525.php), citing the no. 2 of SNCF R^seau, the speed limit for this bridge is 176 km/hr. They also quote Guillame Pepy as saying that these tests are run with some of the safety systems turned off.

There's a schematic diagram of this flying junction at http://florent.brisou.pagesperso-orange.fr/LGVEE2.pdf (http://florent.brisou.pagesperso-orange.fr/LGVEE2.pdf), down in the bottom right corner.

Very Sad.  Lets wait until the investigation.  My suspicion is that they were doing something that they would not be doing in normal passenger service.

With kids on board.

I'm not sure what to think about that.

Presumably, test trains are not regarded as risky. If they were adults wouldn't be on them either (certainly not 49 of them).  Rail engineers are not, in general  reckless people and these are supposed to be safe, controlled tests.  I don't think that there are categories of tests that have a level of risk that is acceptable for 40 odd adults but unacceptable for the children of those adults.  If there is a derailment risk then the test is not safe for anyone to be on board.

You can be sure that kids will be banned in future which is a shame because it will prevent parents sharing their pride in their work with their families.     

It is known These test trains have been running 10% over speed limits with safety equipment off so as it is possible to run at limit plus 10%. The speed at derailment is not confirmed, but the speed limit at location is 160 kph, so plus 10% 176kph. This was around about the 200th test, and all the previous tests ran successfully at the plus 10%.

The train involved was the French version of our NMT train, what kind of staffing does the NMT run with?

The 5 still missing in the above report the BBC are now saying are accounted for and weren't actually on the train, but others may have been due to the "extra guests".

An update on the casualties, supposed to be final apart from four people on the danger list in hospital:
There were 53 people on board, of whom four were minors.
Eleven were killed, but none of the children.
All of the rest were classed as "injured", either slightly or now in a stable condition.

As to the risk of tests, in practice that's a matter of rules. This line was not yet signed off as ready for use, so the rules assume there is a risk that should be assessed technically and revised as tests progress. Only staff needed should be present, just as a standard precaution. SNCF house rules would allow invited guests on trains (not necessarily test trains) once the tests are all done but before all the paperwork is finished. That is reasonable enough, but is likely to be challenged once the lawyers get involved.

Normal track measurements can be done on a line open to passenger trains, so the rules for NMT would presumably be those that apply to it as a workplace. This train was a standard TGV with half the carriages refitted with test equipment, so arguably it was not all workplace in that sense. A new train was used as nothing older is built to run at full line speed, and it will presumably be converted back again. I imagine routine track geometry tests will be done with a "normal" test train, as it does not need full speed to do that out of hours.

Thanks for the extra info.   Presumably, TGVs in normal use have an automatic system to prevent the train going faster than the permitted line speed.  Presumably also, when you do the 10% overspeed test you disable or override that automatic system (or you do it on track before the lineside part of that system is working) .  Presumably, that then allows for human error to result in you exceeding the linespeed by more than 10%.

There's several things there we don't know - obviously you can allow a defined overspeed in several ways, only one of which is to not engage either of the safety control systems fitted. The driver survived and has been interviewed, and says he was sticking to the 176 km/hr testing speed limit. But remember, "only" 176 km/hr is still 110 mi/hr, easily enough to smash a train to pieces.

Grayrigg was about 100 mph for a comparison as to a modern train derailment in Britain under similar speeds.

The TGV was running wrong line, Looking on maps and images of the location, the wrong line looks to have a sharper curve than the normal line. The classic and high speed lines run opposite sides, and the location of crash was where two single lines fly over each other from the high speed line to swap onto the classic line to normally get to the opposite side.

The driver has been interviewed and claimed he was sticking to the limits set of 176kph.

One report I saw said this was the first run on this line in this direction at speed, though I can't see why that would be so important.

SNCF have been saying they can't confirm the actual speed as the gendarmes have the data recorders under lock and key as evidence. I still don't understand why one of the signalling systems wasn't tracking the train, even if it was not enforcing speed limits. The test phase is supposed to test the signalling too.

We were half-promised a first statement of facts from SNCF today, but examining the recorders only started this afternoon, so this first statement is now due tomorrow. I'm sure they will also have examined the track.

If you want a wider view of the site and the track, SNCF have made several videos of overflights of the "LGV Est European" while it was being built; there are links in this page (http://www.lgv-est.com/mediatheque/videos/). The earlier ones give a good impression of how much activity, machinery etc. it takes to build the whole line and its structures pretty well at once. The 2013 one even shows the TBM just emerged from the second bore of the Saverne tunnel.

The latest, which was done last month and so is this the most relevant, is here (https://youtu.be/EqXKffwDIAE). It flies West-East, so the Vendenheim junction is at the end (and the helicopter was overtaken by the test train, wrong road again, 2 minutes in). Saturday's run was on the right-hand track and derailed just before the canal bridge.

We now have that statement from SNCF - and it's quite clear what the cause was: excessive speed. At the point of derailment, the recorded speed was 243 km/hr as against a normal limit of 160 km/hr. For this run, that is described as late braking; 1 km too late. (More later, when I get home.)

Is this line left hand running throughout? Reason for asking is that I've read elsewhere that the norm is right hand running in the d^partements that border Germany (Moselle, Alsace?).

Are LGVs left hand running within France regardless?

Yes, the LGV follows the normal French rule of the road up to the connector at Vendenheim. However, all lines are built and signalled for bidirectional use, hence the need to test them in both directions.

The Germans swapped most of the lines in Alsace-Lorraine over to right-hand running very soon after 1870, though the line from Strasbourg down to Basel had always been that way round. When the French came to reintegrate the A-L network with the French in 1918, swapping back would have been expensive, and in operational terms it was never a big issue, so they didn't. A bit later some of the main lines were equipped with grade-separated "leapfrog" crossovers close to the old border, such as the one at Xouaxange near Sarrebourg (and yes - that's a genuine French spelling, barbaric though it looks). At least one line built since then, like the LGV, runs on the left and has a flyover crossing (or two, I think) just before Metz at Frescaty.

The press conference today (http://www.sncf.com/fr/Presse/fil-info/deraillement-eckwersheim956874) adds only a little to description of the accident. The initial technical report (http://www.sncf.com/ressources/rapport_denquete_deraillement_rame_dessai_814521_sur_le_raccordement_de_la_lgv_ee_le_14_novembre_2015.pdf) does describe how the train derailed; essentially it toppled over centrifugally on the curve (of only 945 m radius). Calculations give a speed of 220-240 km/hr for this to happen, and that the outer ends of the carriage rake and inner ends of the power cars should lift first. That is what happened, and the coupling at this point broke early on leaving the carriages to run off the track and straight over the canal, on their sides. The rear power car never made it over the canal, and most of the (shared) bogies were eventually ripped off.

The leading power car hit the bridge parapet and ended up in three main parts - the body crossed the bridge, scraping along the parapet/girder, and slid down the bank. The rear bogie destroyed the end of the parapet and was embedded in the end of the girder. The transformer went over the side of the bridge, and its oil burning was the source of the fires visible in the pictures.

Given that, it is remarkable that so many people survived, and in particular those in the leading power car.

The press conference, and the full statement, list four sets of actions:

• Suspend all testing at increasing speed on LGVs, and tighten both the rules on not letting outsiders onto test trains and their enforcement.
  
• Conduct a review of the testing methods for new lines, and the implications of this accident for the management of human factors, to report within six months.
  
• Certain individuals will be subject to immediate suspension and disciplinary procedures.
  
• A wider review of operational (not just testing) safety management and human factors is already underway, will take the review of this accident into account.

Obviously much more investigation is needed - this "immediate" report took only four days - and one area where this is true is this question: who was in the driving cab? The report says there were seven - four drivers and test managers and three others. Both of those numbers are surprising; does it need that many drivers? and doesn't driving a train at over normal line speed on the fastest line in Europe call for undivided attention?

The four staff were a driver, a second driver (who was watching), a test manager (also meant to be supervising), and a "pilot" working for Systra (the SNCF/RATP in-house engineering consultancy). Guillaume Pepy's prepared words included a rather oddly-worded section; something about ill-judged behaviour of individuals both in the driving cab and between it and the train.

All in all, it's obvious they did know speed was the cause from the start. Probably they wanted to check against the recordings before going public on that. So Pepy's insistence that this could never happen in service because of the train control systems was based on that knowledge, rather than being reassuring hyperbole. His saying the cause was unknown, which looked incompatible with that assurance, was a legal fiction (i.e. we know but need to know what the prosecutors know).

I've found a report that gives some more detail about the distribution of the casualties. I think we can assume that none of those on-board had any kind of seat belt, in which case the protection given by the body shells was pretty good.

Of the eleven people who died, none was in the front power car (and the report said that most of them were standing). That vehicle slid to a halt over 150 m, an average deceleration of 1.5 g, which helps to explain the survival rate. However, it hit the parapet very hard, and turned on its side, so I still think that survival was surprising.

Five of those killed were Systra engineering staff, and four worked for SNCF; the other two were outsiders. The highest fatality rate was in the last but one carriage, and from those figures it looks like that was a laboratory vehicle.

Sixteen people are still in hospital; two of them are still in a critical state.

So, are we now sure that no children were among the dead?  :-\

Yes, the figures are now given as definitive. However, the identity of the two still on the danger list has not been made public. You will understand that in this kind of accident, with carriages sliding along on their sides, it was always going to be difficult to count the number of those killed and no longer inside.

I would like to thank Stuving for finding and posting all of the relevant information concerning this very tragic incident.

And just to add a little bit to it, this article has a graph showing the train speeds involved:
http://www.railjournal.com/index.php/europe/late-braking-blamed-for-tgv-est-derailment.html?channel=537

The identity of some of the victims is now known. Of the five Systra staff, four held senior positions either for the trials or for the LGV Est as a whole. The wife of one of those was also killed, as was another female guest of a Systra employee who survived.

One of the four SNCF staff killed was Alain Cuccaroni, director of the LGV Est project for RFF and SNCF R^seau since its construction was approved in 1998.

Yesterday's report does contain pictures of the site and the wreckage, much of which can be followed without reading the text. The reason for the high death toll in the last but one carriage is visible in one of the pictures; it went across the canal sideways and hit the far bank upside down.

Thanks for the mention, SandTEngineer, obviously I hope it is of interest. In this case it's not just to draw comparisons, which applies in any serious accident - the "could it happen here" question - but also because what would ordinarily be the lead news item, and for several days in France, has been pushed down or out of the running order. "Interesting times" indeed, in the worst possible way.

SNCF have posted full English transcript of Thursday's press conference (http://www.sncf.com/ressources/sncf_press_conference_english_version.pdf). I'd add a couple of notes to it:

These SNCF internal auditors are not accountants, they look at safety, project management, and other operational matters.

The French Plateforme means the trackbed or formation, but has been unidiomatically translated as platform.

Their version of Guillaume Pepy's sentence that I commented on before reads:

I have read something about the test manager in the driver's cab being distracted by a long conversation on the intercom, though that doesn't sound very serious. Perhaps it was, and we will hear more about it later.

I also found this interesting (actually somewhat alarming):
https://ressources.data.sncf.com/explore/dataset/incidents-securite/?tab=table&sort=date

I'm inclined to agree, though it's hard to judge how alarming this list of accidents and incidents is without something similar from the British railway network to compare it with.

A lot of the incidents are minor, so a total of over 700 during the last year (i.e. 2 per day) may or may not be very high. The fact that more than 30% of them are graded "ESR", meaning there was a potential danger to life, is what I find troubling.

BEA-TT issued a short interim report (http://www.bea-tt.developpement-durable.gouv.fr/IMG/pdf/Note_Etape_BEATT_2015-16.pdf) (really just a short note) in February, which I didn't spot. It starts with a restatement of the facts based on on-board recordings.

• The accident occurred where the speed limit comes down from 320 km/h in stages to 160 km/h for the curve that runs over the bridge and joins the old line towards Strasbourg. The test run was meant to happen at +10% overspeed, but the actual target speeds were rounded up a bit. Remember that the speed must be reduced before a lower speed limit (PSR), which is what ERTMS (or the equivalent SNCF systems) does. More than that, it compares speed during the deceleration with the nominal speed profile and adjusts the braking to ensure it hits that new PSR at (or just below) its speed. The test was driven with ERTMS off, so the driver needed to do that by eye, based on the plan he had been given. Two other drivers were there to "help" him.
  
  1: normal PSR V320, x1.1 = 352, test planned at 360
  
  2: from km 398.77: normal PSR V300, x1.1 = 330, test planned at 330, actual run 331.2 km/h; planned braking for next PSR at km 400.7.
  
  3: from km 403.809: normal PSR V160, x1.1 = 176, test planned at 176, actual run 265 km/h. Derailment occurred at 243 km/h versus a design figure of 235 km/h.
  
  
• They go on to say that the braking point left very little room for error, and that on that run braking started twelve seconds after that point (1.1 km at 330 km/h).  Those figures imply a deceleration of 0.97 m/s/s over 3109 m, or 1.50 m/s/s over 2009 m. Even the higher figure is within the range of emergency braking in good conditions, at least at lower speeds. However, at this high speed the power flux required may not be feasible, so maximum deceleration would be lower - that depends on the design of these trains. The reasons for the delay in braking were still under investigation.
  
  
• Their final point is to question whether these overspeed tests are still needed, or really a survival from an earlier age. A modern instrumented trails train should pick up track defects at much lower speeds, so the overspeed test adds very little - except a big safety issue in how the trains is to be driven. Ever higher train speeds make the result of getting this safety regime wrong more severe as well.
  
  They note that the trials staff seemed very concerned to run every part of the track at the full V+10%, where possible. That is why the braking was scheduled as late as possible; in effect this practice turns the maximum speed (plus 10%!) into a minimum to be achieved.
  
  They referred this question to other parts of the industry (including the safety regulator EPSF).

In today's Le Parisien (and Aujourd'hui en France, its provincial version) there is a report (http://www.leparisien.fr/economie/tgv-est-un-essai-a-risque-mene-trois-jours-avant-le-drame-25-07-2016-5992143.php) based on unpublished information from the enquiry (i.e. the customary leak to the press). This says that:

• On the same run three days earlier, the train ran too fast as the driver had difficulty following the trials plan. There was a video made that shows the several drivers and trials managers debating where to start braking, and missing some of the speed targets by a substantial margin (e.g. 298 km/h rather than 230 - but not at a formal speed limit point). What's really of concern is the implication that they were working this out as they went along, rather than sticking to a plan or using some technical help (like a modified ERTMS).
  
  
• The report makes much of the fact that the trials plan had been modified to raise the top overspeed to 360 km/h in place of 352 km/hr. That speed, of course, was not implicated in the accident.
  
  
• The "cadre transport traction" (senior driver, supervising the actual driver), who was standing behind the driver, admitted telling him to brake 1 km later just before the accident. He said this was based on having managed easily on the last run (though I think that was run on the "down" line, while the accident happened when running "wrong road" on the "up" line*). That bears out the concerns in the interim note about being too concerned to reach the full overspeed wherever possible.

(*French railway tracks are not up or down - they are even or odd. I have no idea why, nor how anyone remembers which is which.)

Thank you for your detailed posts, stuving, and for your helpful and reasoned explanations.  :)

My own schoolboy French certainly isn't up to it.  ::)

Le Parisien / Aujourd'hui en France (http://www.leparisien.fr/economie/tgv-est-des-erreurs-de-calcul-a-l-origine-du-deraillement-31-10-2016-6270525.php) have got hold of an interim judicial (prosecutors') report on this accident, which adds a little more detail to the earlier explanations. The train did brake too late, for a combination of reasons all related to the conduct of these particularly risky overspeed trials.

The correct braking point would have been km 400.2, but the trials plan had decreed km 401 (the initial BEA-TT report said 400.7), and the senior driver overseeing the run changed that to km 402 on the day. Since the speed monitoring and automatic braking systems were turned off, this offhand way of fixing the speeds is the real culprit.

They have other examples, such as where earlier runs went faster than specified but no-one bothered to investigate why, and one of those was the previous run over this section. Elsewhere, the speed was worked out based on the dynamic limit (i.e. when the train overturns) rather than the signalling system's limit (which determines speeds in service).

Apparently those directly involved on the day are facing prosecution, while SNCF and their trials subsidiary Systra may do too.

BEA-TT has published the final report (http://www.bea-tt.developpement-durable.gouv.fr/rapport-final-r256.html) on the accident to a TGV test train on 14th November 2015 at Eckwersheim.  BEA-TT don't seem to do English summaries of their reports any more, even for those important ones that would justify it.

So far there has only been a short note from BEA-TT, and some longer reports from SNCF. However, the narrative of the accident has appeared in various leaks to the press, reported above. The final report tells the same story, though in much more detail.

The recommendations look rather anodyne, and in measured language that I find hard to translate exacltly. Most of them produce the response "you mean they really weren't doing that?". Roughly, they are:

  Recommendation R1 (SNCF Réseau and EPSF - the rail safety body)

Conclude the review of the documents and regulations that govern proving trials for new high-speed lines, including the UIC in this.

  Recommendation R3 (SNCF Mobilités, and any other train operators involved)

To provide drivers for test trains, set up a process to select and suitably train them.

  Recommendation R4 (SNCF Mobilités, and any other train operators involved)

Provide tools to support trials teams planning braking. [For overspeed tests - remember the KVB and equivalent train protection systems were disabled for this.]

Give a clear definition of who does what in conducting the test runs, with respect to driving the train.

  Recommendation R5 (Systra, and any other potential trials managers)

Review the reference documents for trials conduct based on what has been learnt from this accident, in particular to ensure:

• that the manager of the trial has suitable qualifations and experience
• that the timetable allows enough time for preparation and debriefing for each run
• that the manager of the trial and the driver manager coordinate their work by direct contact
• that the role of "pilot" is redefined so he is not the communications link between the manager of the trial and the driver manager
• that a realistic division of functions between the manager of the trial and the driver manager is defined

  Recommendation R6 (SNCF Mobilités, and any other train operators involved)

Install recorders for any telephone link between the drivers' cab and the trials team, and for sounds within the cab.

How the driving function is shared among the on-board team was important in the accident. It is also very much based on the SNCF way of doing things, and not likely to match anyone else's. The functions and names of the four key actors perhaps need a bit of explaining.

1. The Chef d'essais bord (my manager of the trial) is in charge on the train. Back at the office, during planing and similar work, he will be part of a team, but during the trial he's in the cab and is The Boss.

2. The Cadre Transport Traction (my driver manager) directs the driver, based on the agreed "script", and is responsible for safety on board and during planning.

3. The pilote traction (my "pilot") provides detailed knowledge of the line, its speed restrictions, etc. On board he monitors whether driving actions happen in the right place, and watches for signals. He can also direct the driver.

4. The conducteur (driver) drives the train, watching the line and the instruments, but not needing to read the script and think about trials issues as well.

5. There is also a chef de train (train manager) who looks after the doors, and whom to allow into the train and where, as directed by the manager of the trial.

According to reports today (initially from F3 Grand-Est (http://france3-regions.francetvinfo.fr/grand-est/alsace/exclu-accident-mortel-du-tgv-eckwersheim-sncf-societe-systra-reconnues-responsables-1356895.html)) the enquête judiciaire has been finished, and the prosecutors' office has told the families of the results. While the specific details quoted look a bit garbled (they don't tally with the BEA-TT report), the key point is that both SNCF and Systra are almost certain to be prosecuted. That is hardly a surprise, given their lack of clear direction from above. Three employees of the two companies have already been charged with something answering to manslaughter.

As expected, Systra has been mis en examen (roughly charged) and SNCF is expected to follow in a few days. That allows for formal questioning, though how cautioning works with a company I'm not sure.

Finally, with all the urgency we've come to expect of the French legal system, the trial of several bits of SNCF, including Systra, began today. Here from RFI: (https://www.rfi.fr/en/france/20240304-trial-of-deadly-2015-high-speed-train-crash-opens-in-paris-tgv-transport)

Including the driver in this trial seems particularly harsh, given what he's been through and how responsibility was described in the report as lying mainly with others. No doubt there would have been others but they wel killed in the accident.

Mind you, I don't think (without thoroughly revising the subject) the points picked out in that piece are the key ones. They may of course be present in a much longer legal text.

That's a bit misleading; all three of the people who collectively drove the train are on trial. The reports are not using the same terminology as the BEA-TT report (based on SNCF's documents) either. Those three were the hands-on driver, a driver manager who told him what to do and when, and a Systra engineer who handed over the "script" to the other two and then was meant to monitor the trial and answer technical questions.