The latest developments in technology have proven capable of some remarkable achievements across all walks of life. For example, not so long ago, mastering the ancient Chinese game of Go was considered out of reach for machines – with the number of possible permutations exceeding the number of atoms in the universe. A ‘brute force’ approach, which may have worked for the relatively ‘simple’ challenge of chess, simply isn’t possible, forcing the machines to have to think like a human player. Once the stuff of science fiction, it recently became a reality with AlphaGo, the machine that taught itself how to play and then mastered the game.
Other skills that come naturally to humans but have been challenging for a software solution are gradually being mastered – image recognition is a prime example. Nowadays, should you wish to, it is easy to set your favourite search engine the task of finding pictures of, say, cats, and in return you receive a vast array of images of precisely that.
With sufficient training material, software algorithms can learn the key features that allow them to identify correctly the target subject matter in completely new images – not exactly the apocalyptic vision of machines rising up to challenge their human overlords, but a significant development in applying a level of interpretation that, until recently, has been the preserve of sentient beings.
Similar techniques enable self-driving cars to be aware enough of their surroundings to recognise threats and interpret the roadside instructions that we humans take for granted.
Bitcoins and augmented reality
Technology has given us the ability to achieve more in other fields too. New, secure cryptocurrencies, such as BitCoin, have arisen, underpinned by technology that is able to ensure the integrity of transactions without the need for a central banking authority. Instead, a distributed system allows parties with a mutual distrust of each other to cooperate to validate and secure transactions such that these can be trusted implicitly.
Meanwhile, when we want to escape the real world and relax, or at least immerse ourselves in a different kind of stressful environment, it has become perfectly normal to slip on a virtual reality headset and inhabit a different world altogether, which might be completely made up or an enhanced version of real life. Or, why not do away with the headset, and just augment reality on your mobile phone while running around chasing fictional creatures?
So, what’s the link between BitCoin, AlphaGo, self-driving cars, pictures of cats, and virtual reality headsets? And, more to the point, what do they have to do with a feature supposedly all about the more familiar territory of railway signalling?
Well, one thing all of these have in common is that they rely on advanced technologies that are being evaluated for potential railway applications, and, as we move into the next control period for investment, it is likely that some or all of these will start to make an impact on the way we design, build, operate and maintain signalling systems and other assets.
Adoption of these technologies may not be immediately apparent to the traveller going about their journey, but they will, nevertheless, contribute to continuing improvements in safety, financial efficiency and reduced disruption, contributing to improvements that customers will notice.
Managing data using Blockchain
Cryptocurrency relies on a highly secure means of ensuring the safety, integrity and balance of a financial ledger, using an underlying technology known as blockchain. At first glance, that doesn’t sound like anything that would improve railway signalling.
But this technology can be used in all sorts of areas outside of finance. Fundamentally it is about managing blocks of data, and that’s something that we deal with a lot when we’re designing and maintaining our assets.
As we move into an increasingly digital, data-driven world, there’s a growing need to rethink the ways we manage the information we need for designing and maintaining the railway. Traditional paper records, and even electronic versions of these, have data management processes that do not lend themselves to controlling the elements of what is often referred to as a ‘digital twin’, a model that integrates information about the whole network, not just discrete parts of it.
Blockchain, along with other similar technologies, offers potential in managing this data model with secure transactions being used to isolate and securely modify assets, much simplifying the overlapping areas that require careful manual controls with our current file-based approach.
This means that we can maintain a single digital model that represents the state of the whole railway at any given time, while designers are preparing the modifications for renewals and enhancement works within the same model.
This, in turn, provides a more reliable and well-maintained model which reduces the amount of work required to validate it before future changes – especially when coupled with the automated surveying capabilities described below.
So, while the next regulatory period is unlikely to be funded by mining bitcoins, it’s certainly highly likely we’ll be harvesting some of the technology used!
One of the most exciting and challenging areas is where machine learning and artificial intelligence can be brought into use in automating simple and repetitive jobs – those activities that tend to involve a lot of time and effort and tie up skilled designers in performing relatively menial tasks.
For the design and delivery of signalling renewals and enhancements to become as efficient as possible, with a limited pool of skilled talent, automation of the everyday, time-consuming activities is an obvious benefit – it doesn’t reduce the need for skilled staff but means that their skills can be focussed on the activities that those skills are most suited for, while computational power is targeted at the repetitive and straightforward elements.
So what are those areas we need to focus on?
To start with, every resignalling project involves a huge amount of manual correlation of the existing infrastructure, to make sure that the records we have are accurate and we are starting from a correct design viewpoint.
Over the past few years, much effort has gone into refining the methods by which we capture the data that we need to do that – replacing manual surveys with automated laser and video scanning that can, far more quickly and reliably, build a record of the infrastructure at the time.
But the images obtained don’t tell us what is where, they just show us what it looks like. It still requires an engineer to view the footage and tag all the items of interest, identifying what and exactly where they are.
Intelligent machine vision has been used for some time to tackle the challenge of inspecting thousands of miles of rail for defects. This is a problem area that the technology available excels at – the ‘normal’ view of a rail from above is fairly consistent, and so identifying where the captured image deviates from the normal is ever-improving, but fundamentally understood.
Finding signalling (and other) assets automatically from miles of video footage is a rather different prospect. The problem here is that the background image, against which a range of different types of signal must be identified regardless of configuration, type or alignment, is ever changing.
The problem moves from one of finding an abnormal element in an otherwise normal image, to finding an abnormal element in an otherwise random image!
So, given that, what technologies are already achieving aspects of this? Firstly, consider the driverless car. One of the primary safety aspects of such an autonomous vehicle is that it needs to identify an obstacle, possibly a moving one, in its field of vision. This is very much in line with our own challenges because it is using advanced image processing to identify something in the way of the vehicle, an abnormal object against a background that is ever changing.
However, what that object actually is doesn’t matter as much as knowing it is there in order to react to its presence.
Take a more trivial example – the cat photographs. Searching for “cat” results in countless feline matches based on recognition of the content, rather than stored metadata about each picture.
However, as undeniably powerful as this is (at least, if you extend your search beyond the urgent need to find pictures of every single cat ever uploaded), if your search only returns 90 per cent of cat pictures from the unimaginable number of images on the whole of the web, it’s unlikely to be a major problem.
On the other hand, if we can’t be sure we will recognise absolutely every signalling asset from an automated survey, we still need to have someone checking through the entire footage to make sure, otherwise we could miss something that will be a costly error to rectify.
Is finding pictures on the internet a folly that has no place in “serious” engineering? Of course not, it’s a relatively flippant example of the application of technology that has potential to change the way we approach everyday tasks, but we need to work on our own application needs to address the pertinent engineering challenges.
Could AlphaGo have the answer? Well, that’s another strand of artificial intelligence that has definite potential. Building a software solution that, quite literally, taught itself how to play the game, was ground-breaking.
Given the number of possible permutations, a brute force approach to solving potential outcomes beyond a few moves was simply impossible.
What was achieved was a demonstration that machines could learn how to solve specific problems without understanding all the possible outcomes in the first place. This is very much like the problem we’re trying to solve – we need our ‘machine’ to learn how to find signals, and other assets, regardless of the many variations in the position, shape and alignment of these objects and the infinite variety of backgrounds that they might be found against.
By comparison, virtual reality headsets feel tangibly close to regular use – in fact they have been trialled in a number of areas including our own signal sighting tools. (Pictured above and below) Of all the technologies discussed so far, this is the one that will probably be most apparent in the short term, at least to those designing and maintaining the railway.
As an office-based technology for viewing the details of the railway in a safe environment, its use is assured, offering huge safety benefits, through reducing the need for staff to access the live railway, and generating savings of both cost and time.
Applications beyond that need more consideration. There are potential parallels between the heads-up displays used in fighter jets and providing ‘hands-free’ access to extra information while undertaking maintenance activities on the railway. After all, both are in a live operational environment and a fighter pilot needs an immediate and accurate display of information to a greater extent than any likely railway application, save perhaps the driver of a train, so adopting similar technology doesn’t therefore seem unreasonable.
But that pilot is strapped into a seat, whereas railway workers are usually trying to work, and move around, in a live railway environment where slips, trips and falls are an ever-present danger, not to mention keeping out of the way of passing trains. That introduces different risks that have to be managed appropriately.
So, in summary, there are some impressive technologies out there that are providing great opportunities to do things differently, and it’s likely that all of them will be playing a role in the future of signalling design and across many other aspects of railway engineering.
Technology, it seems, moves on faster than ever, and, while the railways have always been an outlet for innovation in engineering, it is just as important to stop and look at the world around us too, and make the most of all that the best innovative technology has to offer.
This article was written by David Shipman, innovations engineering manager, Signalling Innovations Group, Network Rail IP Signalling.
Read more: Ethical engineering