It is too early to say exactly why the recent Spanish train crash occurred. What we do know is that it occurred on a modern high speed route, with up-to-date rolling stock and with an experienced driver at the controls. The public is asking ‘How could something like this happen?’ Here at Heriot-Watt University we have been working on novel ways to use black-box data to help us answer questions like these, and the results are surprising. While it is tempting for the media to blame the driver we have to ask ourselves how far this really takes us? Do we want to solve the problem as stated, or do we want to deal with the real underlying issues?
We are in a slightly strange position. European rail travel is among the safest in the world but at the same time as risk performance has been improving, significantly more people are travelling by train. The UK rail network is fast approaching its historical peak but on a network that is half the size it used to be. What we have, then, are more people travelling on more trains, operating more closely together, on a smaller network, faster. The fact that the safety statistics have continued to improve, despite these big increases in operational intensity, is a major achievement. Unfortunately there is a slight problem, and it concerns the types of risks we now have left-over from the ones we have been able to tackle successfully so far.
Many of the operational risks faced by the industry have, at their core, a prominent human element. To clarify, these are incidents where there is no significant technical fault, no deliberate or wilful violation of procedures, with well-qualified personnel, motivated to avoid an incident, in an environment with every conceivable technical countermeasure to help them. Yet an accident still happens. Take those that have happened on the East Coast Main Line’s Morpeth curve. There have been three which involved over-speeding trains becoming derailed, one in 1969, another in 1984 and yet another in 1992. There were important differences between them, not least that an AWS warning was provided on the approach to the curve by the time the last one occurred. Yet an accident still happened. The problem as stated is put down to driver error, and we find ourselves asking the same question: ‘How could this have happened?’ The real underlying issue, the reason why accidents like these sometimes persist despite seemingly common-sense engineering interventions, lies in a scientific field of study called Human Factors.
Hands up who would like to board a train in London bound for Edinburgh, one that was fully automatic and had no driver? Not many I imagine. The reasons we like a human at the front of our trains – because they are flexible and adaptable, able to deal with unexpected events and keep our trains running on time – are the same as those which occasionally lead to errors. The 1999 accident at Winslow illustrates both sides of this coin perfectly. The precipitating event was a train passing a signal at danger but it was a driver of a following train who managed to turn a 110mph rear-end collision into a significantly less injurious 50mph one. For every human error there are many more heroic recoveries.
Errors seldom arise from some wilful or deliberate intention to disobey the rules. In most cases the human behaviour makes perfect sense to the person in that situation and context because of our underlying human psychology and the way we make sense of the world around us, a world that can sometimes mislead us into behaving the opposite way from what is required (think of all the times you hit send before attaching a file to an email).
Something to emerge from the Spanish rail accident has been the media attention devoted to the train’s black box recorder. Here we face another paradox. Since their mandatory introduction in 2002, On Train Data Recorders (OTDR’s) have coincided with year-on-year improvements in safety. This means they are rarely used for their original purpose simply because accidents don’t happen very often. What we have is a kind of pyramid. At the tip are exceedingly rare incidents such as those in Spain, while at the base we have the mountain of data collected from all those millions of normal journeys where very little of interest seems to be happening. Or so it might seem. The aviation sector was faced with exactly the same problem in the 1970’s. In the dispassionate terminology of the risk field, an accident represents a ‘lagging indicator’ and all we can do is learn from it. What we really want, though, are ‘leading indicators’, things that tell us before an accident happens so that we have time to do something about it. The UK aviation sector innovated an approach called Flight Data Monitoring, which uses black box data to do exactly this. Here the opportunity was recognised to make proactive use of all the data being collected from those countless ‘normal’ journeys. The bigger the data, the more subtle the trends that could be detected which, if they were allowed to continue, might eventually lead to something more serious. So you have a cycle. A continuous living database of black box data, the detection of trends within it, counter-measures, and monitoring to ensure the countermeasures aren’t causing more problems than they solve. What sort of countermeasures exactly? Perhaps a very minor procedural change might be all it takes to arrest a trend? Maybe an adjustment to the infrastructure or equipment? Maybe greater emphasis to a particular subtlety in training? In any case it isn’t really about big traumatic step-changes and it’s certainly not about punitive measures directed at drivers. Data monitoring is about constantly helping the industry evolve towards its desired safety and performance goals.
The rail industry is now at a point where this experience and insight can be brought to bear to inspire a similar form of ‘Rail data monitoring’. This is what we have been examining here at Heriot-Watt University.
Predicting where risks rise
Believe it or not there is still a fly in the ointment. Even despite all this, human factors incidents still percolate to the top to surprise us, even with mature flight data monitoring techniques. That is what our research has been specifically targeted towards. Here in the academic sector we have scientific methods that can accept OTDR data as an input and tell us, as an output, where human factors risks might be increasing. In other words, what we have been developing are various types of ‘human factors leading indicators’ which we can derive from all that routine data being collected from OTDR devices. A simple example is the so-called vigilance decrement. We know that after about 15 minutes of not having to do much, reaction times start to slow down. We can examine black box data for instances where long periods of monotony are broken by periods of sudden high activity. Psychological research tells us what sorts of errors might become more likely if we put people in this situation: large-scale black box data tells us exactly where such situations might be occurring in real-life.
The UK is currently a world leader in flight data monitoring and there is a major opportunity for it to become a world leader in rail data monitoring too. By doing so, we can go beyond the headlines and get to the root of how something like this could happen and stop it before it does.
Dr Guy Walker is a lecturer at Heriot-Watt University’s Institute for Infrastructure and Environment.