[justneedaname]

This reminds me of the very first project that I worked on, a warning system for rail trackside workers. There had been numerous case studies of near misses, injuries and even fatalities.

The system currently in place at the time was, unbelievably, two people (in the case of a bi-directional line) stood downline within earshot of the main crew. When they saw a train approaching they would blow whistles and wave a flag, the workers would then move out of the way until the train passed. Yeah I also couldn't believe that such an archaic system was still in use - this was in 2019 mind.

The company in charge of managing the railway lines reached out to our company and a few others to have us tender on a new design to help protect workers and reduce near misses. Our research led us to an existing system developed by a company in Switzerland which we essentially planned to modify for our national network, as there are differences in how railway lines are signalled across different regions. It consisted of units that could be placed periodically downline of the work site and would alarm when a train was approaching by use of real-time train location data.

The main issue we faced though was how to ensure an accurate reading that gave enough time to vacate the line whilst not being excessive, as research suggested workers may believe it to be a false positive if nothing approached after a couple of minutes. To understand why this is a difficult problem it first helps to understand how traffic within a railway line is managed.

The railway network is split into what is known as blocks, these are discrete sections of track separated by axle counters. Without a train the two sections of track are electronically separate, when one passes over a circuit is completed and the train's position can know be known to that exact location at that exact time. However these readings are discreet, with resolution of the trains position only being as good as the number of axle counters present on the line. This results in some tricky estimating of when "impact" will happen. Train speed is another metric that can be used in conjunction but again you only know the speed read at the last axle counter, anything could have happened between the last reading and "now".

In the end our solution was to assume maximum line speed and warn when this would be within 30s of the worksite. We created a demo that worked flawlessly and the client was visibly impressed. However they then wanted a proposal, cost and everything else for the next phase within 2 weeks - so we had to pull out as we weren't able to produce it. This was a real shame for me as I look back on this project with fond memories, one of the few projects where we were essentially left to figure it out. Already in my short career (<5y) I've been fortunate enough to work on some interesting projects and gain interesting stories to tell...