Rather than tack more onto the post about the Air France crash, the Washington Post is now reporting what I speculated on almost verbatim:
What I had typed:
*** Washington D.C. update ***
6 killed today as one Metro train plows into another that is stopped in a station. Media reports are blaming a failure of the signaling system, rather than the operator not noticing the impending crash and going into emergency stop.
Just a guess, but the most likely cause of this type of accident is when an engineer overshoots a station, then backs up the train on manual. That puts the lead train into a “block” that is supposed to be empty. The following train doesn’t know that because they already went past the green “all clear” signal. The witness reports all agree that the moving train did not apply brakes at all prior to the crash. I’ve ridden the Metro, and its cab control is very thorough – the only thing the operator does is to close the doors. The train controls the acceleration and braking. It’s very plausible that the operator could have been playing video games on her cell phone and letting the train make all the decisions.
A detail as you look at the pictures and read the accounts – you’ll see 4 tracks, with the Metro train on the inner track. The outside tracks are NOT metro – they are the Amtrak line that runs into Union Station. That probably greatly complicated the crash scene and the risk that an Amtrak train could have plowed into the wreckage – fortunately that didn’t happen.
——————————————————————————–
What the Washington Post reported:
“The trains in yesterday’s crash were supposed to be in automatic operation, which means the operators would have been relying on the computerized system to run the trains. The only function required of a train operator during automatic operation is to close the doors after a station stop. Some safety experts said operators can “zone out” during computerized operation because they don’t have to pay as close attention as when they manually run trains.”
———————————————————————————-
Already the cries are coming out for more funding for more technology to fix this system that had two fail safe systems to stop the trains – three if you count the human.
It gets worse… and further supports my theory. There was a bizarre crash on the same Red Line in 2004 involving the same 1000 series (original 1974) cars, with the “anti-climbing” protection failing to work. The NTSB Report can be read here.
The determination was that in this case, the lead train was put into manual mode (it was an empty train). The following train was in the station unloading passengers. The first train rolled backwards for close to half a mile down a hill until it hit the stopped train going 36 mph. The driver had not had much sleep and probably didn’t notice the train was rolling backwards in the dark tunnel.
The operator of the empty train noticed the train starting to roll backwards. He tried to stop the rolling by applying power (like riding the clutch on a car on a hill). But on the older trains that doesn’t stop the train – in fact it can make it roll back faster. The correct action is to *apply the brakes*, and the train will stop. When the train didn’t stop, he didn’t know what to do. The management didn’t know that in manual mode on the old trains the software wouldn’t stop a rollback, so there was no training for this situation.
That’s not what happened yesterday – the area of the crash was a downgrade – there is a curve to the right, with a fence separating the Metro from the Amtrak trains, which does somewhat limit the visibility around the gentle curve, but the top of the train is visible above the fence.
Another interesting detail from the 2004 crash. Non-revenue trains HAVE to be run in manual mode, since they are not in the schedule and you don’t want them automatically stopping in the stations and opening the doors – and have people getting on the train that is just going to the storage yard.
One account says the lead train today had been stopped for 2 or 3 minutes because of “electrical problems”. I will almost guarantee you that the first train was in manual mode trying to override the safety systems – and the train being in manual mode made it “invisible” to the train the crashed into it. Before a train is put into manual mode, the “central control” operations center must notify other trains on the line to come to an immediate stop or operate manually at reduced speeds, or you can get exactly this type of outcome.
One curious report (hard to know if it was accurate) was from a passenger on the moving train… said there was a thump…. and she texted a message on her phone, and then the big crash happened. That would tend to suggest that the following train’s safety system put the train in emergency stop, and then either the operator overrode on manual and/or the train was drifting downhill. The 1000 series cars do not have data recorders on them, and the operator is dead, so it may take a bit to find out what happened.
** added **
A few more random thoughts – the fence and the right hand curve make it more difficult to see if the stopped train was in the south or north bound tracks. Maybe she assumed the train was northbound.
The center position of the speed control is freewheeling – the train neither applies power nor applies the brakes. Since that area has a significant grade heading down, perhaps the train protection system isn’t designed to apply brakes if there is no power being applied – and just like the 2004 accident, the train rolled due to gravity and an inattentive operator….
6/24/09
NTSB’s preliminary analysis says that – at the point of the collision – the moving train was on automatic, train protection was turned on, the emergency stop was depressed, and there is evidence on the wheels and tracks of an emergency stop.
The 2004 accident report contains extensive data of when each train passed each sensor and at what speed, so the NTSB claim that there will be no data about the moving train (because it doesn’t have on-board data recorders) is in error. They also stated that the NTSB rejected the response that the Metro would not retrofit the 1000 series cars is factually wrong – the document (above) published by the NTSB clearly says “answer accepted” on both items. Let the politically motivated lying and finger pointing begin.
6/25/09
Tests yesterday revealed that one section of track failed to detect the presence of a shunt (the shunt creates a bridge between the two rails just as the axle of a rail car does – which is one way the system knows a train is present. An earlier news account stated that this section of track had maintenance performed on it recently. That explains part of the cause, but not all of it. Each train sends out a signal identifying itself – its location and speed that travels on the rails. – that’s how the train protection system knows what to do. Either the lead train was not broadcasting, or there was an electrical break between the blocks. Trains going through a block without being seen should have created “faults” in the monitoring system. Probably they’re going to find the people who worked on the track disconnected the wiring for the safety system and didn’t reconnect it.
*light bulb moment*
The track on the outside of the Metro tracks is not only for Amtrak, it is also Metro Rail, the commuter rail service. If a Metro rail train was going around that curve on the outside track as the subway train was approaching, it could have obscured the moving train’s ability to see the stopped train. Given it was rush hour, that seems very possible. Time to check the train schedules for MetroRail…
Well, that idea doesn’t check out… Maryland Transit doesn’t run any contra-flow trains. The run all of their cars into Union Station in the morning (and have to send 2 trains to Baltimore due to lack of track space in Union Station), and then send all the trains out in the evening. But that track is also an active CSX freight line… (Maryland Transit’s site mentions commuter trains are basically at the mercy of CSX’s freight movements)… so it’s even more likely that a CSX train going inbound could have blocked her view.
WTOP is Reporting that Metro mangement has ordered the 1000 series cars to be used only in the middle of trains (trains are made up of groups of 2 cars – 2,4,6 or 8 cars). By putting them in the middle, the newer trains are controlling the train set, and the 1000 series cars are not exposed to climbing in a collision. They are also rechecking all 3,000 track circuits in the entire system for faults.
——–
Are you sure you want the government running your health care system?
Tags: DC, Metro, NTSC, Public Transit
I thought the automatic train protection system provided continuous signaling, such that if a train backed up into a block the train already in the block would receive a “stop” signal.
The NTSB report on the 1996 metrorail accident distinguishes between the ATP system and the ATS system…it sounded from the report as if the former was continuous in operation but the latter usually downloaded speed targets only at stations.
It’s hard to know for sure. The 2004 accident was in part because the driver was in a 1000 series train, and its software wasn’t current and the Metro management didn’t know that.
When the operator puts the train in manual, it is not under the control of the central control system (thatf would be the ATS system). The ATP system should have alerted the moving train it was getting too close to the train in front, but she has the option to override that. This section of track is one of the longest and straightest sections. My memory (from 20+ years ago) was that 60 mph was the top speed. They’re saying this morning that the impact was probably at about 20 mph.
What turned this from a bad accident to a horrific one is that the 1000 series cars had inadequate protection to prevent the frame of one car going under the other, causing the passenger area to be sheared off. The same thing happened in the 2004 accident, but the NTSB and FRA don’t have the authority to tell them what to do – Metro said they planned to retire the 1000s series soon, and it wasn’t sure it was even possible to retrofit them with anti-climbing protection, so the issue was dropped.
I haven’t heard any reports of the lead train backing up – although the first report of the incident was that it was a collision involving a north bound train…. but nobody on the lead train has mentioned the train backing up.
I wonder if the moving train was a “mixed set” (it was a six car train – 3 sets of 2 cars). Perhaps that is also a factor. They already have the data recorders from the lead train, but that isn’t going to show much. The 1000 series were never fitted with data recorders because of there short remaining life. From the 2004 accident, the trackside sensors apparently log a lot of data, so I’m confident they’ll figure it out soon enough.