Danger Ahead! Special Feature
The Paddington Disaster
Green for Danger - the "Phantom" Signal
the perception of aspect and the information it conveys is crucially important.
Green for Danger
The "Phantom" Signal
by Ray State
On September 29, 1997, about 5:45 p.m. eastern daylight time, eastbound Consolidated Rail Corporation (Conrail) train PIBE-8, consisting of 2 locomotive units and 136 cars, passed a stop and proceed signal at 30 mph at Hummelstown, Pennsylvania and struck the rear locomotive unit of eastbound Conrail train ENS-103, consisting of 5 locomotive units. Each train was crewed by an engineer and a conductor. The conductor on train PIBE-8 sustained fatal injuries in the accident. No other injuries were reported. Damages were estimated at $571,700. The weather conditions were clear, with bright sunlight and a temperature of 65" F. The light was behind the driver. (NTSB Report, 1997)
As rail accidents go this was not earth shattering.
However the cause was singularly significant in consideration of the issues behind signals passed at danger by drivers who have otherwise unblemished records
Where a human is the interface between technology and the control of the event such as the driving of a vehicle the perception is all important. We all are familiar with road traffic lights yet when asked to rank the most dangerous event between the aspects of green, red and no lights because the traffic signals have broken down, the latter is the obvious choice. However, whilst red is clearly a danger signal and usually obeyed without question, the green signal is taken as a clear indication that the driver has the right of passage. The crossing of the junction is then done without looking right or left as it is assumed the other signals (which cannot be seen) are red. Should a set of traffic signals fail to green in both directions the results would be spectacular. Failed lights are seen to be failed and indicate to the driver that something is amiss. Anyone observing vehicles at an intersection with such a failure will note that everyone is creeping about at dead slow speed. The most dangerous indication is green or the perception that it is green.
In the Hummelstown accident, the signal displayed a stop and proceed (red over red signal aspect). This indication calls for the train crew to stop and then proceed at restricted speed.
The engineer stated that both he and the conductor observed and called the signal as "approach medium" (yellow over green signal aspect). An approach medium indication calls for the train crew to proceed to the next signal not exceeding medium speed. The engineer said the train was travelling about 10 mph when he and the conductor observed the signal; the engineer then allowed the train to increase speed to about 30 mph, which would have been an appropriate response to an approach medium signal. He stated that as train PIBE-8 came around the left hand curve, he observed train ENS-103 stopped at CP Tara. The train PIBE-8 engineer put his train into emergency braking but was unable to stop short of train ENS-103.
Post-accident tests revealed that signal 1061E, located about 2 miles west of signal 1081E, was coded to display a stop and proceed signal. The tests also confirmed that the signals were properly wired. Post accident inspection of signal 1061E revealed that the stop and proceed signal was out of focus. Rusty water was found in the signal lens.
When viewed from the track, the signal was partially obscured by tree foliage.
On October 1 1997 National Transportation Safety Board Investigators, with representatives of the Federal Railroad Administration, the Brotherhood of Locomotive Engineers, the United Transportation Union, and Conrail, used a locomotive to replicate the pre-accident events. In sunny conditions, the test locomotive travelled eastbound towards signal 1061E at the same time of day that the incident occurred. Signal 1061E was set to display a stop and proceed signal.
When the test locomotive had moved to within about 1,500 feet of signal 1061E, the signal could not be clearly distinguished by persons on the locomotive. As the locomotive approached the signal more closely, the top aspect of the signal appeared to be yellow and the bottom aspect appeared to be green. Eventually, as the locomotive moved still closer to signal 1061E, the signal aspect could not be distinguished at all. Persons on the test locomotive variously reported seeing yellow, red, and green aspects.
The out-of-focus condition of signal 1061E, in combination with the late afternoon sun shining on the signal face and the water in the lens, probably made the signal aspect appear to the train PIBE-8 train crew to be yellow over green instead of its actual display, which was a red over red aspect. The result was a "phantom signal". A phantom signal is defined by the Association of American Railroads Signal Manual as "an aspect displayed by a light signal, different from the aspect intended, caused by a light from an external source being reflected by the optical system of the signal".
The perception of aspect and the information it conveys is crucially important. The weakness in the current AWS system is that it gives the same warning for single yellow, double yellow and red. Each can be cancelled by the driver which overrides the brake application. It is up to the driver to remember the last aspect and use the evidence of the colour of the signal in front of him to take the appropriate action. This is confused by the possibility of the sequence of signals not being "classic". Normally the successive signals are double yellow, single yellow, red but if the train in front is itself moving then the driver can be faced with a succession of yellows each one requiring him to cancel. It is believed that this repeat and involuntary action was at the root cause of the Wembley accident of 11th October 1984 in which a driver ran through a red signal and had a sidelong collision with a freightliner train. A 1975 accident in South London was blamed on the effect of low sun on the signal aspect.
The perception of the crew to the aspect and the information conveyed has parallels with both Southall and Ladbroke Grove. In the case of Hummelstown the instant the driver and conductor had called the aspect and fixed in their mind the information it conveyed then they would have dismissed any further sighting. The driver at Southall cannot recall the aspects yet somewhere in the sequence of signals leading up to the point of collision he must have acquired some unconscious indication that the line was clear. He must have run at least 3, possibly 4, signals at caution or danger one of them conveyed in his mind that the line was clear - which one and why we may never know. However, it is interesting to speculate that in the Southall accident the train was travelling east on a late September day with the sun low to the south west ( behind the driver). Similarly, the driver of the Bedwyn train had the rising sun behind him as he approached SN109. Confusion as to which line he was on could have set in his mind that he was on the down relief, compounded by the absence of any route direction indication on SN109 (this would have been suppressed because the signal was red) and that the aspect was yellow consistent with a train proceeding in front of him. The warning from the AWS would only confirm this perception.
It is not as if complaints about SN109 had not been made. Both ASLEF and the TOCS had made representations. Whether these complaints refer to siting and visibility of the signal or the effect of low light has not been stated. No doubt this will emerge with time.
Machines can be measured, tested and their performances specified such that their behavioral characteristics are predictable and repeatable. With a human machine we have no such surety in behavioral response. The interpretation and response can vary wildly with fatigue, sickness, stress and distraction. Where we have a man-machine interface in a safety chain then we have risk. It is up to the designers of systems to understand how these risks manifest themselves and design in safeguards to make allowance for the fallible human.
Ray State, Nottingham 12th October 1999
NTSB Railroad Accident Brief Report, ATL97FR020 Rear-end collision/derailment - Conrail - Hummelstown, Pennsyvania, September 29, 1997, Washington
The Paddington Tragedy - Resources
This file last updated: Wednesday, 13-Oct-1999 03:40:19 EDT