Capacity dilemma
With more trains using an increasingly in-demand network, how can operators and planners keep the trains moving? PHILIP HAIGH reports from a conference discussing the problem
With more trains using an increasingly in-demand network, how can operators and planners keep the trains moving?
Suburban traffic on the GWR had always been leisurely by the standards of other lines. So wrote Alan Jackson in his book London’s Termini in 1969. “Uncouth expressions such as ‘rush hours’ were rarely if ever heard on Great Western tongues,” he said.
He records that in 1903 there were only eight suburban trains arriving at Paddington between 0500 and 1000. Over at Liverpool Street, there were 136.
There are more trains today at Brunel’s terminus, but even now some of those arrivals comprise only three-car units. Contrast that with the 12-car trains arriving at many London stations, disgorging their masses onwards to work.
But Paddington is changing. Last year electric operations started with four-car Class 387s running in pairs. From 244 seats on a three-car diesel multiple unit to 888 on an eight-car electric is quite a change.
Further change for Paddington’s commuters is coming with Crossrail’s nine-car Class 345s, that will use a new underground station having left the Great Western relief lines near Royal Oak for new tracks. Crossrail will provide direct links from Thames Valley stations east of Reading to London’s West End and City area.
The continuing rollout of Hitachi Class 800s to Paddington’s long-distance services is leading to service changes on a scale not
witnessed since 1976’s introduction of High Speed Trains. Main line service frequencies change from December 2017’s 12 trains per hour (tph) to 15tph in December 2018, with two more trains every hour for Bristol Temple Meads and an extra hourly departure that alternates between Cheltenham Spa and the West Country. The Relief Lines increase from 8tph in December 2017 to 12tph in December 2019, with Crossrail adding two hourly trains to Heathrow Terminal 5 and two for Reading. Three hourly freight paths remain.
Network Rail expects 0800-0859 demand to rise from around 4,000 passengers on the Relief Lines in 2016 to 16,000 in 2043, while main line demand will rise from 8,000 to 17,000 over the same period (see graph 1). Lines will be full by 2020 in terms of train frequency.
That’s why NR is now looking towards remodelling the station’s approaches. British Rail and Railtrack last did the job in the early 1990s. They opted for a layout that prioritised speed, with turnouts at Ladbroke Grove Junction fit for 50mph as trains switched tracks before approaching on bi-directional lines.
NR Lead Strategic Planner Ben Sturgess revealed the problems this causes today to a conference organised by the Institution of Mechanical Engineers on November 14. Paddington’s throat essentially stretches 1¾ miles out to Ladbroke Grove, and trains crossing (for example) from a high-numbered platform at Paddington to the Down Main can block the throat for some time. The problem Sturgess outlined neatly summarises the challenge a busier railway brings - there are trade-offs between speed and capacity, between capacity and performance, and between performance and speed.
If Paddington is to be full in train terms by 2020, then what of making trains longer as
While planners carefully construct timetables to remove conflicts between trains, the more complex a timetable is the less resilient it will be.
the usual option for increasing capacity? Tricky. Paddington tracks curve sharply at their country end, giving a limit of around 300 metres for trains. Tracks could be extended towards the concourse, but this cuts circulating space and would lead to crowding, Sturgess explained.
So remodelling beckons. This could be complicated because NR suggests building a flying junction at Ladbroke Grove. Back in 2015, it priced the work at between £ 75 million and £175m, and reckoned its value for money to be ‘high’. At the time, it said that an enhanced junction would permit better use of network capability and support better performance. Building a dive-under or flyover replicates the major work recently done by NR at Reading, that has improved performance by reducing conflicting moves.
Sturgess suggested that improvements at Ladbroke Grove would need to be matched by similar grade separation at Didcot East (where trains to and from Oxford diverge from the Great Western Main Line). In 2015, NR rated this plan as poor value for money, but said it couldn’t identify all the benefits it might bring (although it would bring better punctuality).
The twin-track line from Didcot to Swindon would also need improving, he said. This could be done by extending the four-mile loops between Challow and Wantage Road to allow faster trains to overtake slower ones.
Punctuality declines in Paddington’s morning and evening peaks, recovering between them. The drop is more marked for ‘right-time’ than for PPM (Public Performance Measure) targets of within ten minutes for long-distance trains and five minutes for suburban services.
This suggests a link between capacity and performance. But when Sturgess showed the IMechE audience a graph of punctuality against trains per hour for London morning peak arrivals, there was a spread that revealed no correlation (see graph 2). Plotting service frequency against speed showed that as speeds rise, the number of trains per hour falls. Sturgess explained that as speed increases, so must signal spacing, which increases headways (the distance between trains on the same track) and reduces capacity.
He went on to explain that faster trains took less time to clear a signalling block section, but that as speeds increased so signal spacing also had to, but at the square of the speed. The result is a graph that shows for four-aspect signalling the lowest headways occurring at around 30mph (see graph 3). As speeds increase, so does the time taken to recover from perturbation, such as a signal stop. Sturgess said that a stop from 60mph costed 40 seconds, but one from 125mph costed over two minutes (see graph 4).
While planners carefully construct timetables to remove conflicts between trains, the more complex a timetable is the less resilient it will be. Trains graphs typically plot services as sloping lines on axes of time and distance. Trains running at the same speed will have the same slope on the graph. Planners can plot these as close together as signalling allows.
A slower train will have a shallower slope and take up more space on the graph, and so more capacity on the track (see graph 5). Sturgess showed that performance falls as the difference in speed between trains increases. This is traditionally why busy lines had fast lines and slow lines. As the slowest freight trains disappeared, so BR removed extra lines. Today, the difference is returning with
Bratton advised that the best way to minimise interaction was to concentrate on increasing the certainty of trains being in the right place at key points on the network. For this reason, he rejected early running.
the difference between stopping trains and expresses.
That is why trains are often ‘flighted’ from a station. The fastest leaves first and the others follow in order of decreasing speed. There is then a gap in departures until the next flight leaves. Out on the track, the trains within the flight are gradually drawing further apart. The slowest must have arrived somewhere clear of the line before the fastest of the following flight reaches it.
In the days before computers, a train planner would have done all this on paper with a sharp pencil, its line representing the train precisely at any point in distance and time. But trains do not always run precisely - over several days, weeks or months a service might run early or late. Mostly, it will be on time, but its performance will vary (see graph 6).
That planner’s sharp pencil may be better replaced by a thick felt-tip pen, with its wide line giving a better impression of how a train really runs. If, on a five-minute frequency, one train runs two minutes late and the following one three minutes early, then both will be trying to be at the same point. Of course, signalling will slow or stop the following early train, but that action introduces perturbation that can ripple across a timetable.
Mapping how trains perform was the subject tackled by MTR European Operations Director Oliver Bratton. He recognised that trains perform within a range of ‘on-time’, usually on-time, sometimes late, and occasionally early. How a train will actually perform depends on probability, and this makes it hard to define point-to-point timings. If timings cannot be precisely defined, it makes creating timetable models much harder.
MTR mapped performance using train graphs, probability and passenger loadings (variable passenger numbers at stations make it harder to use standard dwell times in models, even for a single station). This helped visualise performance and make the most of space between trains, but even this didn’t account for how one train might interact with another.
One train might delay another, or a third might delay one of them. If both are late, there might be no interaction. Even modelling two trains on a simple section of track with different running times can bring a large range of results in terms of delay if one train is late, particularly if the second train is allowed to go first as a result of the first being delayed.
It led Bratton to ask whether railway companies understand how running one train affects another, and whether they understand which performance is being regulated for. Add more trains, and modelling and visualisation becomes even more complicated. Given this, he advised that the best way to minimise interaction was to concentrate on increasing the certainty of trains being in the right place at key points on the network. For this reason, he rejected early running.
To discover the probability of a train being where it is supposed to be needs good data. This has been the subject of much of Paul Naylor’s work at CPC Project Services.
Naylor has been working with London Underground’s Jubilee Line. The line is twin-track, 22 miles long, and has 25 stations between Stanmore and Stratford. Trains operate under TBTC (transmission-based train control) with ATO (automatic train operation), rather than under drivers’ manual control. Naylor said the system collected 5GB of trainrunning data every day.
He took a typical day ( January 16 2017), with a 30tph service running under 120-second headways. The day had three service delays that totalled 11 minutes. Using a two-minute threshold, delays over the day reached six hours. Plotting trains being delayed and recovering time showed the most gained time at Stanmore, with delays confined to a succession of trains heading west from Baker Street to Stanmore around 0800, just after the first service delay.
Increasing the visibility of delays by reducing the threshold to 60 seconds showed most trains passing Canary Wharf westbound between 1700 and 1830 being delayed. A reduction to ten seconds left Naylor’s chart littered with red and blue dots, representing trains delayed or early. It showed that almost every train was late approaching West Hampstead when heading west. In the other direction, most trains were early at Willesden Green but late at the next station, Kilburn.
These examples come from applying the ten-second delay filter running times between stations, and show that the timetable doesn’t match real running. When it’s applied to dwell times at stations, it shows that peak dwell times are not sufficient to account for the number of passengers in the central section
By finding the problems that cause delays that are otherwise invisible to staff or passengers, an operator could increase capacity or improve performance.
between Baker Street and Canary Wharf. With skilled analysis, the charts could be used to identify where timetables gave insufficient time for a train to enter a reversing siding, or the effect of interlocking release times, or of doors opening too slowly. Naylor cautioned that analysts needed good system and operating knowledge to effectively drill into the detail of delays, but with this they could find problems at junctions and sidings, power faults or poorly performing trains. By finding the problems that cause delays that are otherwise invisible to staff or passengers, an operator could increase capacity or improve performance.
Compared with many NR routes, or even some of LU’s lines such as its sub-surface network of Metropolitan, Hammersmith and City and Circle Lines, the Jubilee is a simple network. Yet it produces a large amount of data every day. Scale this upwards to more complicated lines and the data will increase massively. Even so, Naylor clearly showed that examining it is the key to solving many problems.
Network Rail Capacity Planning Director Chris Rowley provided the conference with a different view of performance. He showed what the company had found from flying a drone above Windmill Bridge Junction in South London one afternoon, when there was no disruption. It revealed that between 1630 and 1815, 39 trains were forced to stop at the junction.
Windmill Bridge is complex. It effectively includes Selhurst, Cottage, Norwood Fork and Gloucester Road Junctions. It permits trains heading north from East Croydon to diverge for London Bridge or Victoria. It offers the same option for trains from West Croydon. It Great Western Railway 387133 waits at Paddington on March 3 2017. JACK BOSKETT/ RAIL. also switches the layout of tracks from being paired by speed to being paired by direction.
There is considerable potential for delays to spread from a junction this complicated. Rowley used the phrase “spread like wildfire”, and explained that it was very difficult to add time to services or create firebreaks.
Windmill Bridge goes some way to explaining why performance for the lead operators on NR’s South East route has been consistently between two and ten percentage points below similar suburban operators on Wessex and Anglia, according to Rowley. Plenty of people have tried to improve performance, and there have been several emergency reviews. “There’s clearly a fundamental problem down there,” said Rowley, before revealing what he thought it was.
He showed his IMechE audience a graph of the top 19 busiest Network Rail junctions (see graph 7). Twelve sit on Sussex and the wider South East route, with 27 trains crossing per hour at the busiest - Ewer Street between Waterloo East and London Bridge.
NR has plans to remodel Windmill Bridge Junction. Revealed in 2015, they would add two platforms to East Croydon, and remodel its layout to provide eight rather than six through tracks, which then feed into a revised Windmill Bridge that features new bridges. A note attached to the plan stresses that it is only indicative and will be subject to detailed design, funding, operational analysis and
planning powers.
The lines through Windmill Bridge Junction will soon be linked to the East Coast Main Line via Canal Tunnels near St Pancras. According to one senior railwayman, ECML performance can be “carnage” on a bad day. Current Moving Annual Average performance for Virgin Trains East Coast is 84.7%, with only Hull Trains (another ECML user, 79.9%) and Govia Thameslink Railway (GTR, 79.5%) below it. GTR includes Thameslink, Southern and Gatwick Express services over Windmill Bridge and Great Northern trains on the ECML.
How delays might be transmitted from one side of Canal Tunnels to the other remains to be seen. Thameslink trains running through the core under central London will be running in automatic mode. Leaving the core, they will fan outwards to several destinations much as water leaves a hose spray. The trick to punctuality is funnelling the spray into the hose at the other end of the core.
Crossrail has the same problem. In the central section, its Class 345s will run automatically under CBTC (communicationsbased train control), but Oliver Bratton noted that this system does not look beyond its central boundaries to see how trains are approaching. It can’t adjust their running as they approach, to ensure they are in the correct order and on time. Thus Bratton admitted that no one knew whether CBTC would transmit delays from either side of the central section or absorb them. Thameslink faces the same situation.
Bratton suggested that not all problems can be solved by computers. Models of planetary motion produce results that astronomers know are wrong but good enough. Results from rail models were similarly good enough, but on a busier network their flaws begin to be exposed. Current rail computer models take a long time to build, and (as Rowley noted) their results can often come too late to change anything. He argued that expert judgement remained the most important area, but that this needed an accurate description of the changes being proposed.
The long timescales involved make it difficult to try different things to find what works best. Increasing links between different lines and routes makes modelling one in isolation much harder.
This is why Network Rail is looking to create a new model using the huge amount of train running data it collects. NR System Operator Managing Director Jo Kaye admits she doesn’t know how this will work out, but the company is talking to academics to discover what might be done.
The idea of a model on which large or small changes can be tested across the network is powerful. It qualifies as another ‘holy grail’ of train planning. LU’s Jubilee Line experience shows what can be done with data, but NR’s ambition for its model is vast - not just in the geography of its network, but also in the daily and hourly variations in train running caused by events inside and outside the railway’s control. Tweaking trains around Brighton could ripple toward Edinburgh via Thameslink; Birmingham could find itself linked to Chelmsford through CrossCountry services to Reading, Crossrail through London and onto the Great Eastern Main Line.
At the heart of NR’s problems is that a busier network brings stiffer links between areas where once empty space on timetable graphs made links more elastic. It’s akin to operating a complex mechanical machine, full of levers and gears. You need fine tolerances between parts for the machine to run smoothly, but the variables in train performance mean that your tolerances are wide and sloppy and so the machine jams.
If links between timetables are becoming tighter, so the railway must consider links between projects, particularly if Network Rail is to continue delivering upgrade schemes as well as maintaining and renewing the railway it already has.
GWR Managing Director Mark Hopwood told the IMechE conference that projects affecting his trains were developed in isolation from others. Earlier delivery of the double-track upgrade from Bicester to Oxford for trains from Marylebone could have helped passengers disrupted by work between Paddington, Reading and Didcot, he argued, even though he ruefully added that the Bicester project had caused GWR some commercial pain.
Hopwood argued that adding trains did not always mean a reduction in performance. The West Coast Main Line is running more reliably now than before Virgin introduced Class 390s, while the North London Line has better performance and more trains. Both lines were subject to extensive upgrade work. The challenge today, Hopwood reckoned, was that so many upgrades were taking place at the same time
Although the Mechanicals’ conference carried the title Performance or Capacity: Either
or Both? it’s clear that passengers and freight forwarders want both. Rail companies must plan to deliver both, but must be prepared to explain where compromises lie.
This was a point NR’s Chris Rowley made strongly. He said the railway industry was not “done to”. It proposes changes and must be able to tell funders, particularly governments, where trade-offs lie.
Money can explain some of these compromises. Adding capacity allows railway companies to sell more of their product - whether that’s seats, container space or track paths.
Passenger companies now have to hand money back to passengers when trains are late. The Department for Transport’s ‘Delay Repay’ scheme entails a full refund for a train an hour or more late, and half your money back for delays over 30 minutes. Recent franchise awards have reduced the threshold to 15 minutes. This should create a strong commercial link between performance and capacity.
Rail companies should design both aspects into their timetables from the start. And as IMechE Railway Division Chairman Richard McClean said at the end of November 14’s conference, they must then concentrate on basic operations, maintenance and renewals, and deliver their product locally with proper grass-roots understanding.
If links between timetables are becoming tighter, so the railway must consider links between projects, particularly if Network Rail is to continue delivering upgrade schemes as well as maintaining and renewing the railway it already has.