From London to Doha: How to leverage signaller workload benchmarks for new signalling projects
New automatic train control systems are transforming the tube by providing increased capacity, improving reliability, and making journeys faster for commuters. The new automated systems offer improvements behind the scenes too because they can be operated with fewer people than the traditional signalling systems. Increased automation of traditional signaller tasks, and concepts such as flexible areas of control, can make predicting the right number of signallers difficult. However there is a growing opportunity to leverage the data gathered from previous signalling upgrade projects as a benchmark.
I’ve co-authored a paper on the subject with Mark Lowten – Head of Human Factors at Thales GTS – for the ‘6th International Rail Human Factors Conference’. Our paper compares the benchmarks we've used to predict the number of signallers needed for two recent projects; my work on the London Underground Four Lines Modernisation project upgrading the Sub-Surface Railway, and Mark's work for the new 4 line metro system in Doha, Qatar – both of which make use of the Thales CBTC system.
Here's a brief overview of what you can expect from the paper and presentation;
Size matters. Though how you measure 'size' matters more.
One challenge I ran into was how to measure the size of a signalling area. Track length is part of the story, but that alone doesn't come close to capturing the features of a railway that contribute to a Signallers' workload. The 'Workload Hotspots' approach I developed objectively quantifies a total of 17 railway features, such as the number of reversing locations, depot interfaces and complicated junctions to measure the 'size' of a signalling area.
Like for Like
The next step of the process is to gather the workload hotspot data for the benchmark.
It's important to remember that benchmarks are most useful when there are like-for-like comparisons. Look out for similarity in terms of the level of automation being provided as well as the similarity with the staff operating model in terms of the operator roles and responsibilities. These are the main reasons I used benchmark data from the recently upgraded Northern Line - which uses both very similar technology and staff operating models.
If a similar benchmark can't be found then it's probably best to use a ‘Day In The Life Of’ workshop instead.
Don't worry – you don't need to create a total ‘workload score'
When I started developing ways to objectively quantify workload hotspots, I was asked by some stakeholders if I could then add up the individual workload hotspots to provide a total ‘workload score' for each signalling area.
This is really difficult for two reasons;
There's often no natural way to add up workload hotspots because they use a variety of different units (Km, quantities, frequency etc).
Workload hotspots are not created equal, so they would need to be weighted relative to their importance before adding them up. There's simply isn't enough data available to weight them confidently*. Even if you could, the relative importance of each workload hotspot depends on the scenario – which means that there isn't a single, meaningful weighting to be derived anyway.
Instead, I found that using a data table to compare benchmark signalling areas with the proposed signalling areas made it really easy to compare each type of hotspot, like with like. This approach was certainly simple, but it was also transparent which I think helped get stakeholder buy-in because I wasn’t using “magic formulas”!
I also made maps to illustrate the geographic location of the workload hotspots and the size of the signalling areas of control. This really helped stakeholders to visualise the data, and made it easy to mark up changes.
You should try benchmarking too!
The paper concludes that using benchmarks to predict the number of signallers is really useful, and could be used as an alternative to ‘Day In The Life Of’ workshop - hopefully after reading this you will be inspired to try it yourself!
Remember to follow these steps;
Identify the railway features that contribute to operator workload. The level of automation being provided will have a big impact on what to include.
Objectively quantify the features - I called these 'workload hotspots'.
Create a benchmark using data from similar railways. Consider how similar the benchmark technology and operator roles and responsibilities are – the closer the better.
Compare like with like - don't bother aggregating the workload hotspots into a single score!
Involve your stakeholders throughout. Input from subject matter experts is incredibly useful when developing a benchmark for verifying data, for making close judgement calls, and balancing the areas of control so they make sense operationally.
BUT you do need to be cautious if there are differences between the benchmark and the new system being developed – a ‘Day In The Life Of’ workshop will help in these situations.
Lastly, as with any prediction, it makes good sense to validate the result using follow up assessments.
Related posts:
Event Report: Hammersmith Service Control Centre Design and Integration at Tech Soc Rail Human Factors Conference 2015; Digital Display Design for Engineering Vehicles Rail Human Factors Conference 2013; The Observatron
*I tried to consider how a serious mathematician might approach the problem of coming up with a total workload score based on the number of workload hotspots. A bit like those really serious mathematicians who come up with a formula for the “perfect” pizza. That sort of mathematician might gather the workload hotspot data from lots and lots of other similar railways, and then use multiple linear regression to derive the relationship between the types of Workload Hotspots and the number of signalling areas. And to be confident in the answer, the mathematician would need to consider if there is enough data available. A good rule of thumb is 10-20 data points for each dimension used in the regression (in this case a dimension would be each type of 'workload hotspot'). Based on the 17 workload Hotspots I’ve identified, I would need the data from 170 to 340 similar railways to use multiple linear regression and create a total workload score! This is unfortunately not possible yet, but I think the real insight here is this; if that's how much data you need to have any chance of mathematically understanding the sum relationship between workload hotspots and the size of signalling areas, then I am highly suspicious of any model that purports to achieve something similar via a workshop with a panel of experts.
