Norfolk Southern Train Derailment: NTSB Preliminary Report

Here is the U.S. National Transportation Safety Board preliminary report, “Norfolk Southern Railway Train Derailment with Subsequent Hazardous Material Release and Fires” [PDF]. The track along which the train travelled prior to the accident was equipped with wayside defect detectors which monitor, among other items, bearing temperatures of axles of passing trains (“Hot Bearing Detector” or HBD). The report states:

On the Fort Wayne Line of the Keystone Division, NS has equipped their rail network with HBD systems to assess the temperature conditions of wheel bearings while en route. The function of the HBD is to detect overheated bearings and provide audible real-time warnings to train crews. Train 32N passed three HBD systems on its trip before the derailment. At MP 79.9, the suspect bearing from the 23rd car had a recorded temperature of 38°F above ambient temperature. When train 32N passed the next HBD, at MP 69.01, the bearing’s recorded temperature was 103°F above ambient. The third HBD, at MP 49.81, recorded the suspect bearing’s temperature at 253°F above ambient.

After the train stopped, the crew observed fire and smoke and notified the Cleveland East dispatcher of a possible derailment. With dispatcher authorization, the crew applied handbrakes to the two railcars at the head of the train, uncoupled the head-end locomotives, and moved the locomotives about 1 mile from the uncoupled railcars. Responders arrived at the derailment site and began response efforts.


Theoretically, the emergency brake should operate by simultaneously braking all cars to prevent accordioning. Accordioning should only occur if a lead car or locomotive braked and trailing ones did not.


Here is a video explaining how railway defect detectors work.


That’s why they have started to put electrically operated air releases onto railroad cars – because the engine can release the air, but it take some time to decompress all the way through the train, so that the front cars are braking, but the rear cars are still moving with all their intertia.


That’s inexcusable. When you pass an HBD, it tells you either that there is a hotbox somewhere on your train N axles back, or it says “no defects found”. If you, as an engineer, pass an HBD, and don’t hear the report, you should stop your train and let the dispatcher know.


In 2019, trains in the US traveled 777 million train-kilometers and experienced 1,338 derailments. [0.0058Mm per derailment]

The same year trains in the EU traveled 4.5 billion train-kilometers and experienced 73 derailments. [0.61Mm per derailment]

Japan: 2 billion train-kilometers and 9 derailments. [2.2Mm per derailment]

via Scott Santens: "In 2019, trains in the US traveled 777 million tr…" - (haven’t verified the numbers, but seems plausible)


I don’t get the train-km per derailment numbers from the figures cited in the linked post. From Fourmilab Units Calculator:

(777 million Train km) / (1338 Derailment) = 0.58071749 (million Train km) / Derailment

(4.5 billion Train km) / (73 Derailment) = 61.643836 (million Train km) / Derailment

(2 billion Train km) / (9 Derailment) = 222.22222 (million Train km) / Derailment


I toyed with the megameter (Mm) unit, converting from the underlying social media post.

Are the derailment numbers comparable? Rail traffic in EU & Japan is primarily passenger service, whereas rail traffic in the US is overwhelmingly freight. It would be reasonable for the standards for passenger service to be higher.


In Switzerland, it’s difficult to directly compare passenger and freight traffic because passenger traffic is quoted in train-kilometres and seat-kilometres, while freight is measured by net tonne-kilometres. A sense of the mix is given by the number of locomotives assigned to each service: 322 for passenger and 221 for freight. This is for the Swiss Federal Railways and does not include regional railways such as BLS, which is 55.8% owned by the canton of Berne and 21.7% by the Confederation (but not the Federal Railways).


Another example of such biases is this road safety display - that neglects to incorporate the variability in the miles driven (or ridden) across geographies:

1 Like

Here is a table from the Junior Woodchucks’ Guidebook of road deaths, which I have displayed sorted in descending order of number of deaths per billion vehicle-kilometres. The countries shown are the only ones for which such figures are published—it is likely there are many countries worse than Mexico, number 1 here, for which no data are available.


I was astonished to see that Israel comes in between Japan and France on this measure. That certainly isn’t what I’d have guessed from crossing the street there.


Even in the above spreadsheet, there’s an interesting contrast between low population density countries (Norway/Sweden) and higher population density countries (Belgium/Hong Kong) - but also between the Latin cultures in Europe (Belgium) and Teutonic ones (Netherlands).

Speaking of Latin (including Israel) vs Teutonic, there’s a certain order in chaos, as long as everyone knows the rules:


Switzerland is not a low population density country. It is 71st in the world with 211 people/km², comparable to Italy (197) and Germany (234). Norway is low density, 215th in the world, with 17 people/km². Mexico is relatively low density, rank 154, with 65 people/km². By comparison, the U.S. ranks 186th with 35 people/km².

For highway death statistics, for some countries such as Norway and Switzerland, I think you’d need to look at “effective population density” taking into account that substantial parts of both countries are essentially uninhabited with few roads, and most of the population and highways concentrated into a portion of the territory.