Not Just GWR Then.......

[whole diary]

[SandTEngineer]

Confirms then that the level crossing sequence was correctly initiated, cancelled when the train appeared to reach the inner, second train coming strike-in point, and then reinitiated again when it reached the strike-out point for the wrong direction running.  That looks like an infrastructure and not a train issue to me.

[stuving]

Confirms then that the level crossing sequence was correctly initiated, cancelled when the train appeared to reach the inner, second train coming strike-in point, and then reinitiated again when it reached the strike-out point for the wrong direction running.  That looks like an infrastructure and not a train issue to me.

Most of the comments I have read say that this crossing (and others on the Cromer line) do not use treadles, but have predictors based on track circuits. Hence all the talk about wheel contamination etc. I have located an article in Rail Engineer (Issue 152 - June 2017) that describes these:

Level-crossing predictor
With automatic level crossings, the disadvantage of a fixed strike-in point is that the actual warning time given to the user may vary significantly according to the speed of the train. To overcome this issue, a level-crossing predictor was first introduced between Norwich and Cromer when the line was resignalled in 2000. The GETS Harmon HXP-3 uses audio frequency track circuits to detect an approaching train, and the rate of change of the inductance of the rails is used to determine its speed and hence calculate the trigger moment to provide a constant warning time for each train. Another similar product is the Siemens Automation Wayguard WESTex GCP4000.

In another article, the reliability of these systems is questioned:

Train arrival prediction systems to dependably predict the time of a train’s arrival at a level crossing have been available for some time, but performance is variable and not consistent enough to provide ‘safe to cross’ information. The Network Rail strategy is to continue to develop a means of accurately and consistently predicting the arrival of trains at crossings that overcomes the limitations of current systems.

However, I'm not sure how up to date that is, as the same article also says:

Investigating the interlocking of highway road traffic lights with the signalling system at areas where drivers running red lights is a known problem, or at high volume road junctions near to level crossings, is under way.

... and I know where those have been used since well before 2017!

[stuving]

Of course, the definitive place to look for information on a crossing is the sectional appendix. For the record, that says of the Norwich Road crossing (and two others south of Salhouse station):
It's an AHBC^▸ (Automatic Half Barrier Crossing)-X, so bidirectional on both tracks
It is "not provided with treadles", but nothing is said about what is provided, oddly
Signalling is TCB^▸ (Track Circuit Block), but this is one of several sections that are within axle counter limits (with gaps between).

And that, pretty much, is all it says.

[SandTEngineer]

Hmm. The point I was making was that the crossing appears to have had its initial strike-in operate correctly (barriers down and road lights operating).  At some point the strike-in was lost and the sequence cancelled (barriers raise and road lights extinghuish).  It then re-initiated around the wrong direction strike-out point. I never mentioned treadles to back up that operation, its entirely reliant on track circuit operation.

[stuving]

Hmm. The point I was making was that the crossing appears to have had its initial strike-in operate correctly (barriers down and road lights operating).  At some point the strike-in was lost and the sequence cancelled (barriers raise and road lights extinghuish).  It then re-initiated around the wrong direction strike-out point. I never mentioned treadles to back up that operation, its entirely reliant on track circuit operation.

Point taken - I ought perhaps to have said what I did pick up on. I read "strike-in point" as being a fixed point, whether a treadle or a box emulating one. But mainly that set me off looking for more descriptions of this kind of box, to help me or anyone else knowing nothing about them.

From those descriptions, the predictor has to work out the train's position, and from that its speed and direction, to decide the right activation point. The design is well over 20 years old (and American) but still new enough to have a lot of algorithmic stuff in it to do that prediction process. In which case it could - and arguably should - be doing some sanity checks on its data. For example, if it sees a train vanish from its track circuit while approaching, it looks odd to me that the control box should conclude it's really not there any more and vehicles can be allowed to cross.

Is being American relevant? This kind of wrong-side failure is usually associated with modern, light, smooth-running, short DMUs^▸ (Diesel Multiple Unit). To that might be added running on newish well-aligned rails, since most American track is nothing like that (nor their trains neither), and most of it sees only or mainly freight trains. So I could believe that the operating logic of the firmware in these boxes wasn't originally designed to cope with poor wheel-rail electrical contact, and if it has been upgraded that was only just enough to make it work with the track and trains presented to it.

One thing I've not found made clear is the status of the track circuits used by the predictors. I think they are an overlay on the signalling - not used by it, and with their frequency picked to avoid any interference even in AC track circuit areas.

[SandTEngineer]

Interesting advance information from Roger Ford of MODERN RAILWAYS:

INFORMED SOURCES e-Preview February 2020

Crossings modified for Class 755 operation

Last month’s Modern Railways reported that the Rail Accident Investigation Branch (RAIB^▸ (Rail Accident Investigation Branch)) was investigating a very, very near miss at an Automatic Half Barrier (AHB) level crossing on the Norwich-Sheringham line.

 Here’s the RAIB description of what happened.

“At about 19:53 hours on Sunday 24 November 2019, a 4-coach class 755 passenger train, operating the 19:45 Norwich to Sheringham service, was approaching Norwich Road automatic half barrier level crossing, to the north-east of Norwich.  The crossing barriers were in the lowered position until the train, travelling at about 45 mph (72 km/h), was about 200 metres from the crossing.  The barriers then lifted, the level crossing warning lights went out and cars began to cross the railway.  The train driver applied the train’s emergency brake and sounded its warning horn, but the train was unable to stop before reaching the crossing.  No road vehicles were struck but a car passed in front of the train around a quarter of a second before the train went over the crossing”.

After privatisation, one of Railtrack’s priorities was to obtain low cost signalling, suitable for replacing manually controlled mechanical boxes and crossings.  In November 1997, two pilot schemes were let, one of which was the Norwich-Sheringham route.

Harmon of the USA would provide the single processor interlocking and other hardware.  The contract was let through its recently acquired subsidiary Vaughan Systems of Ware, now renamed Vaughan Harmon Systems Ltd.  Three existing automatic crossings would be modernised and three others automated.

With automatic level crossings, the disadvantage of a fixed ‘strike-in’ point, where the train is detected starting the crossing closure sequence, is that the warning time given to the user may be longer than necessary for slow trains.

Harmon’s predictor uses track circuits to detect an approaching train and determines its speed.  The speed is used to calculate the ‘strike in’ point to provide a constant warning time. 

Harmon’s predictor system was fitted to the six AHB between Norwich and Cromer and entered service in 2000.  According to ORR^» (Office of Rail and Road formerly Office of Rail Regulation - about), there has not been any previous accident history on the crossing in the 20 years since the predictor system entered service.

What seems to have happened is that the track circuit lost detection of the approaching Class 755.  The crossing assumed the train had left the section and opened the gates. 

RAIB has identified two key issues for its investigation.

These are the design, implementation and operation of the predictor system, including any effects of rail head contamination due to fallen leaves, plus the design of ‘relevant elements of the Class 755 train and the process for accepting it for use on this route’.

After the incident Network Rail immediately put the barriers under local control and imposed a 20 mile/h Emergency Speed Restriction.  This accounted for much of the service disruption.  The extended running times meant that there was insufficient time to run to Sheringham and pick up the return timetable.  Trains were therefore turned back at Cromer.

It was also decided to fit treadles as a backup to the predictor track circuit at the six AHBs.  This is a reversion to British Rail practice which combined track circuit and treadles.

Norwich Road, Great Plumstead and Rackheath Road crossings had been equipped just over a week after the issue emerged.  The closed period for the gates has also been extended to 98 seconds.  The other three AHB crossings on the route were fitted early in January.

Other routes were also affected by loss of detection of Class 755 multiple units during leaf fall.  The leaf fall period officially ended on 18 December, but testing was still needed to check that Class 755s were being detected correctly on track circuited lines.

In the column I describe the recovery process, which was assisted by some of Network Rail’s ‘intelligent infrastructure’.  By the start of the New Year the infrastructure was fully available with no restrictions on Class 755 operation.  The only exceptions were the 20 mile/h ESR^▸ (Emergency Speed Restriction ) at the three AHBs pending fitment of treadles.

Now we wait for the RAIB report.

[stuving]

We now have the RAIB^▸ (Rail Accident Investigation Branch) report on this incident.

Summary

On 24 November 2019, the barriers at Norwich Road level crossing, near New Rackheath, Norfolk, lifted as a passenger train from Norwich to Sheringham was approaching. Two road vehicles crossed the railway in front of the train, which reached the crossing less than half a second after the second road vehicle was clear.

The investigation found that there was contamination of the railhead in the area caused by leaf-fall and atmospheric conditions. This contamination had not been removed because there were no railhead treatment trains on the Norwich to Sheringham line at weekends. The narrow band on which trains? wheels were running on the contaminated railhead, which was a consequence of the introduction of new trains, left the wheel-rail interface vulnerable to a poor electrical contact in the event of contamination. This caused the level crossing equipment to misinterpret the position of the train, and consequently it opened the crossing to road traffic while the train was closely approaching.
 
RAIB has made three recommendations addressed to Network Rail regarding the planning of autumn railhead treatment, guidance on the introduction of new trains and the configuration control of signalling equipment. RAIB also identified two learning points concerning the investigation of incidents and the signalling design process.

In effect, that confirms the stories circulating after the events, but doesn't really come up with "the answer".

However, there was one more specific issue, not mentioned in the summary - the reset time-out was set at 16 seconds, which was far too short. That's how long the controller waits after the "train detected input" vanishes during an approach to the crossing before it cancels the closure and raises the barrier. The NR^» (Network Rail - home page) (or Railtrack at the time) standard said it should be 120 s,though this controller has a maximum setting of 99 s. This was identified in later installations (Norwich Road having been the initial pilot) and 99 s chosen, so the RAIB recommendation is about applying things learned later to systems already  in use.

RAIB reports are usually very good at clear explanations, sometimes even seeming to go a bit too far. But this one has one area that I find absolutely baffling, and not explained by other data I found earlier. The section on "Background Information" covers how the HXP3 predictor works, and says that the bit of track it uses to detect the train and estimate its position and speed extends to 1610 m from the crossing. There is a shunt across the track at that point that looks like a train to the currents being used, so trains further away don't affect the voltages detected on the rails.
 
But there are three crossings here (Norwich Road, Great Plumstead, and Rackheath Road) all within 1600 m. So if the operating ranges of three controllers overlap, how do they do it? And there are three graphs of voltage vs train position, labelled for the three crossings but "Rail volts seen from Norwich Road". Quite why are they the shapes they are, and what does the origin marker indicate? Nothing tells us. Can anyone else make sense of this?