Chris Harrison
Principal Future Systems Engineer, RSSB
Some 20 years ago, when experts wanted to look at risk on the railway, there was an absence of granular detail. The best they could produce with the information available was a general model of risk that applied to the national network as a whole.
Fast forward a couple of decades and those same experts (older, wiser, and now with a few grey hairs) are, quite astonishingly, able to look at risk narrowed down to 25-metre sections of the track.
These models help operators to take decisions that keep people safe by predicting the effects of change on the network. The picture of current and near-future risk that these models provide has been built up from the data industry shares with us. While successful, this approach can only achieve so much with the amount of data now available. And there is a lot of data.
This is great, as it allows a better understanding of the factors that affect risk. On the flipside, it can make it harder to identify and predict change, particularly if we rely on traditional techniques. Another limitation of our current models is that they take as their starting point the railway as it is today.
Considering all the data we now have available, to look further into the future, for example predicting a certain type of risk in 2040, we need a different way of thinking.
And this is where artificial intelligence — or more precisely, machine learning (ML) — can come to the rescue. ML is a set of tools for making inferences and predictions from data.
ML can take the historical data, as was processed in our traditional modelling, and use it in combination with other very large datasets to work out which variables have the strongest relationships with risk. And when we say very large, we do mean very large. ML can work on combined datasets measured in terabytes.
The ability of ML to work with and make sense of large volumes of data opens up new possibilities to identify multiple and complicated risk factors. These can be used to work out trends that can be projected forward using future rail scenarios to predict risk. It’s an exciting opportunity that rail needs to be able to exploit.
For example, if we’re looking at the derailment risk on a rail line today, ML can help us look further ahead, say 20 or 30 years, and consider how that risk might change based on multiple factors. ML can potentially help us model issues such as climate change, population growth and settlement, and future travel patterns.
It can help us build up a picture of plausible future scenarios, the level of risk and, importantly, help us to plan for risk and mitigate it where necessary. This would help to avoid decisions made now that are undermined by future events or circumstances. A more sophisticated predictive and proactive approach to risk management is what will help improve safety.
AI has been coming in for a lot of stick recently. People are understandably concerned that it could make some roles obsolete and take away jobs. Some are concerned about data privacy and misuse. Others point to its potential for poor-quality output. But there are many different types of AI and many different ways it can be used. For the rail industry, it can be a force for good, enabling us to manage risk far more effectively and look into the future with greater sophistication than before.