The warning system on a train is set off by a magnet between the rails at every signal, and there is no software in it at all: a permanent magnet in the four foot, a sensor under the cab, a horn. The signal at this site came out in 2016 and the magnet with it. The horn still sounds, and a gaussmeter on the sleeper reads a field.
I am a signalling maintenance technician, and the automatic warning system on my section is the part of the job that has not changed since my father did it.
How it works, because the simplicity is the whole of this.
Between the rails, a set distance on the approach to every signal, there is a permanent magnet. Behind it there is an electromagnet that is energised when the signal is clear. A train passing over feels the permanent magnet first, which arms the system, and then the electromagnet. If that one is energised a bell sounds in the cab. If it is not, a horn sounds and the driver has to acknowledge it or the brakes go on.
The permanent magnet is a lump of alloy in a steel case bolted to the sleepers. It has a specified field strength, and we measure it with a gaussmeter on a schedule, and one that is weak gets replaced.
There is no software anywhere in that chain.
At one site on the up line there was a signal until 2016, when the section was resignalled and the signal was taken out.
The magnet came out with it. I know because I signed the work off. The steel case was unbolted from the sleepers and lifted into a road rail vehicle and it went to scrap with the rest of the equipment.
The horn sounds in every cab that passes the site.
Every driver on that route knows it. It is in the route knowledge as a phantom, which is the word drivers use, and they acknowledge it and carry on, because that is what you do with a horn.
The train's own recorder logs an AWS warning at that location on every pass, and an acknowledgement.
What has been checked.
The trains. This is not one unit and it is not one class: it is every train on the route, including a test train from a different operator with a different make of receiver.
The sensor on a train is a coil, and it responds to a magnetic field in the four foot and to nothing else. Not vibration, not heat, not a signal from anywhere.
The track. There is no magnet in it. I have walked it, dug the ballast out between the sleepers at the location, and looked at bare formation.
And I have measured it.
A gaussmeter on the sleeper at the site reads a field of about nought point four millitesla, pointing up, over a patch about the size of the case that used to be bolted there.
Nought point four is a weak AWS magnet. A new one is stronger. A weak one is exactly what you get from a magnet that has been in the track for thirty years, which the one I removed had been.
The sleepers were replaced in 2020 in a renewal. The field reads the same on the new sleepers.
What that means, as far as I am able to say it.
There is a magnetic field in the track, of the right strength, the right polarity and the right size, at the right distance from where a signal used to be, with no magnet there.
I have had a colleague from the electrification side out with a better instrument. He mapped it. His plot shows the field of a bar magnet the size of an AWS magnet at sleeper level, and he asked me twice whether I was sure there was nothing buried, and I dug it out again in front of him.
What is done about it.
The site is in the sectional appendix as a location where a spurious AWS warning will be received, which is a real category, because the system is old and there are a handful of these on the network for reasons nobody has ever run down.
Drivers acknowledge it. It costs nothing and it has never caused an incident.
I raised it in 2021 as a fault against the location, on the grounds that a warning with no signal behind it is a warning that teaches drivers to acknowledge without looking, which is the one habit the system exists to prevent. That is the safety case for doing something about it and it is a good one. The response was that there is no equipment at the location to repair or remove and that the entry in the appendix is the appropriate control.
What I do.
I measure it on the schedule, along with the real ones, and write the figure down, because I cannot think of a reason not to.
The one change is that I do not measure it in the dark.
The gaussmeter test is a probe held on the sleeper for a few seconds, and it is done in a possession, at night, like most of the work.
In February 2023 I was crouched over that spot at about two in the morning with the probe on the sleeper and the reading at nought point four, and a train's warning horn sounded.
Not in a cab. There was no train. The possession was mine and the line was blocked both ways for two miles. The horn sounded out of the dark from the up direction at the pitch and the length an AWS horn sounds in a cab, once, and stopped, and I stood up and walked back to the vehicle and did not put my hand on that sleeper again.