Data and AI in Railway Condition-Based Maintenance

Making Use of Intelligence

Maintaining a railway is hard work. Keeping both rolling stock and tracks in good order demands increasing amounts of time and money. Naturally, it is imperative for rail networks to find the most cost-efficient and effective ways of keeping their networks in good shape. That’s where condition-based maintenance comes in…

But what is condition-based maintenance? In essence, it involves the continual monitoring of track and rolling stock condition, identifying deprecation early and remedying it before it becomes an issue. Inevitably, digital has and continues to play a part in enhancing the effectiveness of this kind of maintenance. Through use of AI and data analytics, condition-based maintenance has come into its own, which – in the context of a rapidly expanding digital railway market (forecast to be valued at $74.8 billion globally by 2024, with a CAGR of 8.4%) – shows just how important these solutions will be to its evolution.

Let’s explore the role of AI and data analytics in condition-based maintenance, how each has opened up new possibilities in the field, and how the intersection between digital and rail may be one of the most significant challenges that network operators have to deal with.

Keeping Track of Data

Big data has many uses in modern railways. Notably, it forms part of the emergence of ‘smart railways’, encompassing interlocking platforms to traffic management systems. But how does it manifest itself in the field of rail maintenance?

Fault detection is no doubt where big data’s impact is most felt. This data can be used to create predictive algorithms which anticipate areas where faults might occur, allowing engineers to anticipate issues and resolve them before they become disruptive. Condition-based maintenance often involves data being drawn from multiple sources, this ultimately being leveraged to proactively target sections of track which are more likely to fail due to their general condition or their susceptibility to failure.

Of course, it’s not as simple as having the data right in front of you and being able to identify track faults with ease. Through techniques such as feature extraction and data imputation, raw data is broken down into manageable chunks, while gaps in the raw data are filled in using accurate estimates. At this point, the retrieved data is compared against the expected data results coming from different parts of the track (also known as control limits). A prognostic model applied to this scenario would then predict the implications of this fault, which would be followed by the decision-making stage and – finally – the presentation of evidence collected through this process.

Big data merely forms one part of a wider network of techniques and tools used in condition-based maintenance, although an integral one. Let’s look at how Artificial Intelligence continues to make these processes far less dependent on the human touch…

The AI Evolution

The use of AI in rail maintenance is not a new phenomenon. In fact, its lifetime as a rail-related tool began in the 1980s, with this iteration being primarily rules-based (i.e. simple rules would be input by humans which the AI system would follow). However, more recent developments in the application of AI on the railways have seen its focus shift towards a machine-learning approach, in which the AI system is ‘taught’ how to identify track faults and maintenance issues proactively and, even, resolve them.

The concept of predictive maintenance is gaining traction in the world of rail. Sensors placed on the track are able to inform engineers of potential faults which may occur, with warnings being communicated between trackside sensors and onboard systems. What’s more, analysing track conditions is just one purpose for which AI-based maintenance can be employed. For instance, SNCF – France’s national rail operator – has used AI for maintenance purposes on everything from onboard air conditioning units to lineside pantographs. Not only does this allow for the real-time identification and resolution of issues, it also reduces costs. It is estimated that efficiency gains of between 10-15% could be made globally through the use of real-time conditions-based maintenance, allowing for this money to be spent on infrastructure investment in other areas.

We’ve explored how both AI and big data will revolutionise maintaining railway networks, but what does the future bear for each condition-based maintenance?

On the Horizon…

It’s clear that the use of AI and big data has only just begun in the rail sector. Initiatives such as the European Union’s ‘Horizon 2020’ programme promise greater investment for the implementation of new technologies on rail networks, as well as support for start-ups driving this innovation. For instance, AI forms a central pillar in the EU-funded OPTIRAIL scheme, which aims to further bring technology to the forefront of fault detection and prevention. In the context of this project, the use of AI will also contribute to a better understanding of the conditions of the network in question, allowing for a longer-term view of how best to prepare for and deal with potential faults.

In the UK, plain line pattern recognition is leveraging the advances brought about by AI to identify track faults remotely. This is carried out through use of cameras positioned on the lineside which compare images taken at two different points in time, with AI algorithms being employed to highlight physical differences in the track between images and alert an engineer controlling the system remotely. This form of Light Detection and Ranging technology can also be employed to prevent geotechnical faults on railway lines. In this instance, Digital Terrain Models are built from imaging produced by the LiDAR technology over time which, through analysis performed by machine learning and cloud technology, can identify geographical threats to certain sections of track.

When it comes to new technologies transforming the way rail maintenance works, be in no doubt the future is now. So, pick the right partner to ride with you down the tracks towards a more intelligent way of keeping networks up-together!